WO2011092865A1 - 物体検出装置及び物体検出方法 - Google Patents
物体検出装置及び物体検出方法 Download PDFInfo
- Publication number
- WO2011092865A1 WO2011092865A1 PCT/JP2010/051355 JP2010051355W WO2011092865A1 WO 2011092865 A1 WO2011092865 A1 WO 2011092865A1 JP 2010051355 W JP2010051355 W JP 2010051355W WO 2011092865 A1 WO2011092865 A1 WO 2011092865A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- strong
- discriminator
- discrimination
- target object
- depicted
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V10/7747—Organisation of the process, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
Definitions
- the present invention relates to an object detection apparatus and an object detection method.
- Patent Document 1 an apparatus and a method for detecting an object based on image information are known (see, for example, Patent Document 1).
- the detection device described in Patent Literature 1 learns information (for example, a feature amount) related to a detection target object (for example, a face) included in image information, and configures a plurality of homogeneous classifiers (homogeneous classifiers) based on the learning result. To do.
- the plurality of homogeneous discriminators are configured such that the detection accuracy of the detection target object varies, for example, by changing the number of feature amounts.
- the apparatus described in Patent Document 1 cascades a plurality of homogeneous discriminators in series so that the detection accuracy of each homogenous discriminator becomes gradually more accurate toward the end of processing. Configure the detector.
- the device described in Patent Document 1 inputs image information of a determination image to a detector, causes the homogeneous discriminators constituting the detector to execute in series, and all the homogeneous discriminators detect a detection target object Only when the detection target object is detected from the determination image information.
- Patent Document 1 determines that one of the plurality of homogeneous discriminators constituting the detector does not include the feature amount of the detection target object in the determination image information, It is determined that the detection target object is not detected from the determination image information without executing processing performed after the homogeneous classifier.
- the apparatus described in Patent Document 1 is configured so that the detection accuracy of the homogenous discriminator constituting the detector becomes gradually more accurate toward the end of the serial coupling, so that the detection process
- the feature quantity discriminated by the homogeneous discriminator tends to increase as the process ends. That is, there is a possibility that the processing speed of each homogeneous classifier decreases as the stage after the series connection. For this reason, in the apparatus described in Patent Document 1, the speed of the object detection process may decrease.
- an object of the present invention is to provide an object detection apparatus and an object detection method capable of speeding up object detection processing.
- an object detection device is an object detection device that detects the target object depicted in a determination image based on a feature amount of the target object learned in advance using a learning image.
- a plurality of weak classifiers that respectively calculate estimated values indicating the possibility that the target object is depicted in the determination image based on the feature amount of the target object, and a plurality of the estimated values,
- a plurality of strong discriminators that discriminate whether or not the target object is depicted in the determination image with different discrimination accuracies, and the strong discriminator that are executed in descending order of discriminating accuracy. If the strong discriminator determines that the target object is not depicted in the determination image, the processing is continued.
- a detector that does not detect the target object without executing the strong discriminator, and the strong discriminator inputs a discrimination result of the strong discriminator having a discrimination accuracy lower than that of the strong discriminator. And determining whether or not the target object is depicted in the determination image based on a plurality of the estimated values and the input determination result.
- the strong discriminator constituting the detector inputs the discrimination result of the strong discriminator having the discrimination accuracy lower than that of the strong discriminator, and the input discrimination result is It is used to determine whether or not the target object is depicted in the determination image. That is, in this object detection device, each strong discriminator does not independently discriminate the target object, but each strong discriminator discriminates the target object using the discrimination results of the other strong discriminators. In this way, the strong discriminator uses the discrimination results of the other strong discriminators, so that each strong discriminator uses a weak discriminant for calculation compared to the case where each strong discriminator independently discriminates the target object. The number of estimated values of the vessel can be reduced.
- the processing speed of each of the strong discriminators can be increased. It becomes possible. As a result, it is possible to speed up the determination of whether or not the target object is depicted in the determination image.
- the strong discriminator may input the discrimination result of the strong discriminator having the second lowest discrimination accuracy than the strong discriminator among the plurality of strong discriminators.
- the strong discriminator is configured to output the determination image based on a weighted voting result using a weight indicating a degree of ease of discrimination of the weak discriminator and an estimated value of the weak discriminator, and an input discrimination result. It may be determined whether or not the target object is depicted.
- the strong discriminator as a discrimination result of the strong discriminator having a discrimination accuracy lower than that of the strong discriminator, includes a weight indicating a discrimination ease of the weak discriminator and an estimated value of the weak discriminator. You may input the result of the used weighted vote.
- a weight corresponding to the discrimination accuracy is assigned to each of the plurality of strong discriminators, and the strong discriminator adds the weight of the input strong discriminator to the input weighted vote result. It may be determined whether or not the target object is depicted in the determination image using the integrated value. In such a configuration, the degree of reflecting the discrimination results of other strong discriminators is changed according to the weight of the strong discriminator as the input source. For this reason, since the discrimination results of other strong discriminators can be appropriately reflected in the self discrimination, it is possible to improve the discrimination accuracy of individual strong discriminators.
- the object detection method includes a plurality of weak classifiers that respectively calculate estimated values indicating the possibility that the target object is depicted in the determination image based on the feature amount of the target object,
- An object detection method for an object detection device comprising: a plurality of strong discriminators that discriminate with different discrimination accuracy whether or not the target object is depicted in the determination image based on the estimated value of An execution step in which the strong classifiers are serially coupled in descending order of discrimination accuracy, and the strong discriminator executes discrimination in ascending order of discrimination accuracy; and the strong discriminator has a lower discrimination accuracy than the strong discriminator.
- the strong discriminator determines that the target object is depicted in the determination image
- the execution step is continued, and the strong discriminator does not depict the target object in the determination image. If it is determined, the execution step is interrupted.
- the strong discriminator may input a discrimination result of the strong discriminator having the second lowest discrimination accuracy than the strong discriminator among the plurality of strong discriminators.
- the strong discriminator uses the weighted voting result using the weight indicating the degree of ease of discrimination of the weak discriminator and the estimated value of the weak discriminator, and the input discrimination result. Based on the determination image, it may be determined whether or not the target object is depicted.
- the strong discriminator has a weight indicating the degree of ease of discrimination of the weak discriminator and the weak discriminator as a discrimination result of the strong discriminator having a discrimination accuracy lower than that of the strong discriminator.
- the result of weighted voting using the estimated value may be input.
- a plurality of strong discriminators are given weights according to the discrimination accuracy, and in the discrimination step, the strong discriminator adds an input source result of the strong discriminator to the result of the weighted vote. Weights may be integrated and it may be determined whether the target object is depicted in the determination image using the integrated value.
- the object detection apparatus and the object detection method which are one embodiment of the present invention, it is possible to increase the speed of the object detection process.
- the object detection device is a device that detects (discriminates and identifies) an object depicted in an image based on image information.
- a personal computer For example, a personal computer, a digital camera, a mobile phone, a PDA (Personal Digital Assistant ) And the like.
- the object detection apparatus learns the characteristics of the detection target object before the detection process, and performs the detection process based on the learned characteristics.
- the target object to be detected is not particularly limited. For example, a human face or the like is used. In the following, a face detection device mounted on a portable terminal having a camera function will be described as an example of the object detection device according to the present invention in consideration of ease of understanding.
- FIG. 1 is a functional block diagram of a mobile terminal 3 including a face detection device 1 according to the present embodiment.
- a mobile terminal 3 shown in FIG. 1 is a mobile terminal carried by a user, for example.
- FIG. 2 is a hardware configuration diagram of the mobile terminal 3.
- the mobile terminal 3 physically includes a main storage device such as a CPU (Central Processing Unit) 100, a ROM (Read Only Memory) 101, and a RAM (Random Access Memory) 102, a camera, a keyboard, and the like.
- the input device 103, the output device 104 such as a display, the auxiliary storage device 105 such as a hard disk, and the like are configured as a normal computer system.
- Each function of the portable terminal 3 and the face detection device 1 described later is configured to load the input device 103 and the output device 104 under the control of the CPU 100 by reading predetermined computer software on hardware such as the CPU 100, the ROM 101, and the RAM 102. This is realized by operating and reading and writing data in the main storage device and the auxiliary storage device 105.
- the face detection apparatus 1 normally includes a main storage device such as the CPU 100, the ROM 101 and the RAM 102, the input device 103, the output device 104, and the auxiliary storage device 105. It may be configured as a computer system.
- the mobile terminal 3 may include a communication module or the like.
- the mobile terminal 3 includes a camera 30, a face detection device 1, an image composition unit 31, and a display unit 32.
- the camera 30 has a function of capturing an image.
- an image sensor or the like is used as the camera 30.
- the camera 30 has a function of outputting the captured image to the face detection device 1 as a determination image.
- the image composition unit 31 has a function of generating a composite image in which a symbol or an enclosure that emphasizes the face portion of the determination image is superimposed on the determination image.
- the display unit 32 has a function of displaying the composite image generated by the image composition unit 31.
- the face detection device 1 includes a determination image input unit 10, an image region division unit 11, and a detector 2.
- the determination image input unit 10 has a function of inputting an image captured by the camera 30 as a determination image.
- the image area dividing unit 11 has a function of dividing the determination image input by the determination image input unit 10 into predetermined areas.
- the image area dividing unit 11 has a function of dividing the image area of the determination image into a plurality of small areas (so-called subwindows) having a predetermined size.
- the sub window may have a rectangular shape or other shapes.
- the positions of the subwindows can be overlapped or not overlapped.
- the image area dividing unit 11 changes the magnification of this subwindow to various sizes. Thereby, the range to be processed in the target image can be changed.
- a conventional method can be adopted as a method of changing the magnification.
- the detector 2 has a function of inputting a subwindow divided by the image region dividing unit 11 and determining whether or not a face as a detection target object is depicted in the subwindow. That is, the detector 2 has a function of detecting a face displayed in the input subwindow. The detector 2 determines whether or not a face is depicted based on image information (such as luminance values) and rectangular features (Rectangleectfeatures) of the subwindow.
- the rectangular feature is a kind of local feature, and for example, a Haar-like feature is used.
- FIG. 3 shows an example of the rectangular feature.
- 3A to 3D show four types of rectangular features 20a to 20d.
- 3A and 3B are for extracting features appearing at the edge of the face, and the rectangular features 20c and 20d shown in FIGS. 3C and 3D. Extracts features that appear in the line of the face.
- Each rectangular feature 20a to 20d is evaluated by the difference between the sum (or average brightness value) of the pixel values (luminance values) in the white area and the sum (or average brightness value) of the pixel values in the black area. For example, in the case of the rectangular feature 20a, the evaluation is based on the difference between the sum of the luminance values in the white region 20a_B and the sum of the luminance values in the black region 20a_A.
- Each rectangular feature 20a to 20d can be applied to an arbitrary position in the sub-window. FIG.
- the detector 2 calculates the difference between the sum of the luminance values in the white region 20b_B of the rectangular feature 20b and the sum of the luminance values in the black region 20b_A. Is calculated.
- the black region 20b_A surrounding the eyeline is often darker than the white region 20b_B surrounding the cheek, nose, and cheeks below.
- the detector 2 learns such human facial features in advance, and determines the discrimination result based on whether or not the calculated difference is greater than a pre-learned threshold.
- the detector 2 includes a plurality of weak discriminators 20n (n: integer) in order to efficiently perform such processing.
- the weak discriminator 20n is a discriminator having a relatively low discrimination ability, and has a function of calculating an estimated value indicating the possibility that a previously learned facial feature is displayed in the sub-window.
- a plurality of weak classifiers 20n are prepared corresponding to the above-described rectangular features, calculate the difference between the sum of the luminance values of the white region and the sum of the luminance values of the black region with respect to the corresponding rectangular features, This is a threshold function that outputs 1 or 0 that is an estimated value based on the magnitude relationship.
- the weak classifier 20n can be expressed by the following Equation 1.
- x is a feature amount
- f j (x) is a function of the weak classifier 20n. That is, f (x) is a function that calculates the difference between the sum of the luminance values of the white region and the sum of the luminance values of the black region using a rectangular feature corresponding to the feature amount x.
- the detector 2 combines the weak discriminators 20n to construct a strong discriminator 21m (m: integer) having a relatively high discrimination accuracy.
- the strong discriminator 21m performs weighted voting based on the plurality of estimated values calculated by the plurality of weak discriminators 20n and the weights assigned to the weak discriminators 20n, and uses the result to display a face in the subwindow. Has a function of determining whether or not is displayed.
- the combination and number of weak classifiers 20n used for weighted voting differ depending on the strong classifier 21m. For this reason, each of the strong discriminators 21m has a different discrimination accuracy.
- the detector 2 discriminates one subwindow with the plurality of strong discriminators 21m, and finally discriminates whether or not a face exists in the subwindow based on the discrimination results of the plurality of strong discriminators 21m.
- the combination of p j ⁇ ⁇ 1,1 ⁇ of the weak discriminator 20n and the threshold T j , the weight of the weak discriminator 20n, and the weak discriminator 20n used by the strong discriminator 21m is, for example, according to the AdaBoost Algorithm. Learned in advance.
- the Adaboost algorithm is one of machine learning methods, and learns so that image information that is difficult to identify by a simple identification method can be identified by a combination of a plurality of simple classifiers.
- a simple classifier that is a base corresponds to the weak classifier 20n.
- the final discriminator in the Adaboost algorithm is the strong discriminator 21m.
- a set of face images in which a face as a detection target object is depicted, a set of non-face images in which a face as a detection target object is not depicted, and a set of weak classifiers 20n are prepared.
- p j and T j of all weak classifiers 20n are provisionally determined using a set of face images and a set of non-face images.
- a weight k j is prepared and initialized for each of the face image and the non-face image. The weight k j indicates the importance in discrimination, and an image having a larger weight k j is an important image that should not be erroneously identified.
- the weak discriminator 20n are optimized to minimize the weighted error, and the weak discriminator 20n using the weak discriminator 20n having the smallest weighted error in the strong discriminator 21m. Adopt as. Thereafter, the weight k j of the face image and the non-face image is updated. Thus, the weight k j is updated every time one weak discriminator 20n is selected. For example, the weight k j of the learning image that cannot be discriminated well by the selected weak classifier 20n is updated so as to increase. For this reason, it becomes easy to select the weak discriminator 20n that can identify an image that has been difficult to identify until now by repeating the above processing.
- the weights assigned to each weak discriminator 20n, strong discriminator 21m is configured for weighting the vote by using the function h j of weak classifiers 20n.
- the strong discriminator 21m represented by the following formula 2 is constructed by Adaboost learning.
- the weak classifiers 20n is h j (x)
- strong discriminator 21m is equivalent to the S 0.
- the weight w j of the weak discriminator 20n means voting power and indicates the degree of discriminability (ie reliability) of the weak discriminator 20n.
- the weight w j is calculated using an error rate based on a weighted error during learning.
- the weight w j of the weak discriminator 20n that determines that a correct face image is almost a face is set large. Further, the weak classifier 20n that hardly determines a correct face image as a face only needs to reverse the determination, and thus the weight w j is set large.
- ⁇ 0 is a value calculated during learning based on the weight w j , for example, and when S 0 is larger than ⁇ 0 , it is determined as a face.
- S 0 is passed through more than A% of the face image for learning, and, until the passing of B less than% of the non-face image for learning, performs selection and additional weak discriminator 20n.
- a and B can be set arbitrarily. By changing A and B, a strong discriminator 21m having different discrimination accuracy can be constructed.
- S 1 which is a strong discriminator 21 m with high discrimination accuracy is learned and generated.
- S 1 is so determined accuracy is better than S 0, A than that in the learning of S 0, B is set strictly.
- a strong discriminator S i (i: natural number) shown in the following Expression 3 is generated.
- the strong discriminator 21m is configured to input the discrimination result of the strong discriminator 21m having the next lowest discrimination accuracy than the strong discriminator 21m.
- W i is a weight set for each S i of strong discriminator 21m.
- the weak discriminator 20n is selected from the beginning and the discrimination accuracy is higher.
- the number of weak classifiers 20n to be selected / added can be reduced as compared with the case where the strong classifier 21m is constructed.
- Detector 2 the S 0 and S i to run serially generated a strong discriminator 21m, linear combination.
- Each strong classifier 21m that is linearly coupled is also called a stage.
- the strong discriminators 21m are connected side by side so that the discrimination accuracy increases as the end of the linear combination.
- the detector 2 receives the sub-window, the detector 2 sequentially executes the strongly combined strong discriminator 21m.
- the detector 2 causes the strong discriminator 21m having the next highest discrimination accuracy to execute a detection process.
- the strong discriminator 21m does not detect a face, the detector 2 Processing after the appearance of the strong discriminator 21m having higher discrimination accuracy than the discriminator 21m is not performed. Further, except for the first stage, each of the strong discriminators 21m inputs the discrimination result of the strong discriminator 21m having the second lowest discriminating accuracy next to the strong discriminator 21m and performs its own detection processing.
- FIG. 5 is a flowchart showing the operation of the face detection apparatus 1.
- the process illustrated in FIG. 5 is executed, for example, at a timing when the camera function of the mobile terminal 3 is turned on, and is repeatedly executed at a predetermined cycle. Note that the processing of S10 to S16 shown in FIG. 5 is executed by the face detection apparatus 1, and the processing of S18 to S22 is executed by the portable terminal 3.
- the determination image input unit 10 first inputs a determination image (S10).
- FIG. 6 shows an example of the determination image F.
- the image area dividing unit 11 generates an integral image of the determination image input in the process of S10 (S12).
- the image region dividing unit 11 scans the determination image F to generate a plurality of subwindows Gn (S14).
- the detector 2 selects one of the generated subwindows Gn and performs face detection processing (S16).
- FIG. 7 the detector 2 is strong discriminator is a S 1 ⁇ S n serially to be executed by the low determination accuracy sequentially 21m.
- S 1 ⁇ S n carries out processing on the same subwindow Gn.
- the detector 2 ends the process at that time, and when it is determined that the face is a face, the process is continued.
- S 2 to Sn are used to input the result of the previous stage and determine whether or not the face. Then, the detector 2 determines that a face is depicted in the subwindow Gn that has passed through all the stages. The detector 2 performs the above process on all the subwindows Gn.
- the process of S16 ends, the process proceeds to a determination process (S18).
- the image composition unit 31 determines whether or not a face has been detected in the process of S16.
- a composite image in which the position of the sub window Gn is emphasized is generated (S20).
- the display unit 32 displays the composite image (S22).
- the determination image is displayed as it is (S22).
- the determination image F is input and divided to generate the subwindow Gn, and whether the subwindow Gn displays a face in each stage is determined based on the result of the previous stage. Is judged.
- the determination result of the preceding stage is inherited by the succeeding stage, it is not necessary to evaluate the subwindow Gn from the beginning in the succeeding stage. Further, the detection accuracy needs to be improved as the number of subsequent stages increases. However, since the determination result of the previous stage is input, the detection accuracy can be improved by adding fewer rectangular features. For this reason, it is possible to suppress an increase in processing time occurring in the subsequent stage.
- a detector for determining the subwindow Gn is, S 1 ⁇ serially to execute S n in ascending order of discrimination accuracy is strong classifiers.
- S 1 to S n function independently, so that the rectangular feature used in each stage becomes more complicated as it becomes later, and as a result, the amount of calculation processing in each stage also becomes later. To increase. Furthermore, even if a strong discriminator up to a certain stage gives a sufficient result, it is rejected by the result of one subsequent stage, so that the detection accuracy may be lowered.
- the strong discriminator 21m constituting the detector 2 inputs the discrimination result of the strong discriminator 21m having a discrimination accuracy lower than that of the strong discriminator 21m. Then, it is determined whether or not the face 40 is depicted in the determination image using the input determination result. That is, in this face detection device 1, each strong discriminator 21m does not discriminate the face 40 independently, but each strong discriminator 21m discriminates the face 40 using the discrimination results of the other strong discriminators 21m. To do. Thus, since the strong discriminator 21m can use the discrimination result of the other strong discriminators 21m, the strong discriminator 21m is compared with the case where each of the strong discriminators 21m independently discriminates the face 40.
- the number of estimated values of the weak classifier 20n used by can be reduced. For this reason, even when the detection accuracy of the strong discriminator 21m constituting the detector 2 is configured to be gradually more accurate toward the end of the serial coupling, the processing speed of each of the strong discriminators 21m is reduced. It becomes possible to speed up. As a result, it is possible to speed up the determination as to whether or not the face 40 is depicted in the determination image. Furthermore, since each strong discriminator is discriminated by reflecting the results of the previous stages, the discriminating information can be discriminated. For this reason, detection accuracy can be improved as a result.
- the detector 2 integrates the weights W i-1 of the input source of the strong discriminator 21m on the result S i-1 of the input weighted vote is integrated It is possible to determine whether or not the face 40 is depicted in the determination image using the obtained value. For this reason, since the discrimination results of the other strong discriminators 21m can be appropriately reflected in their own discrimination, it is possible to improve the discrimination accuracy of the individual strong discriminators 21m.
- the above-described embodiment shows an example of the object detection device according to the present invention.
- the object detection device according to the present invention is not limited to the object detection device according to the embodiment, and the object detection device according to each embodiment may be modified or otherwise changed without changing the gist described in each claim. It may be applied to the above.
- the present invention is not limited to this.
- embodiment mentioned above demonstrated the example which applies the face detection apparatus 1 to the portable terminal 3, it is not restricted to this.
- the example in which the object detection device performs the detection process by inputting the image from the camera 30 has been described, but the input image of the object detection device is not limited to this. For example, it may be an image acquired through communication or an image stored in a storage medium.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims (10)
- 学習画像を用いて予め学習された対象物体の特徴量に基づいて、判定画像に描写された前記対象物体を検出する物体検出装置であって、
前記対象物体の前記特徴量に基づいて、前記判定画像に前記対象物体が描写されている可能性を示す推定値をそれぞれ算出する複数の弱判別器と、
複数の前記推定値に基づいて、前記判定画像に前記対象物体が描写されているか否かをそれぞれ異なる判別精度で判別する複数の強判別器と、
判別精度が低い順に前記強判別器を実行させ、前記強判別器が前記判定画像に前記対象物体が描写されていると判別した場合には処理を継続し、前記強判別器が前記判定画像に前記対象物体が描写されていないと判別した場合には当該強判別器よりも判別精度が高い前記強判別器の実行を行うことなく前記対象物体を未検出とする検出器と、
を備え、
前記強判別器は、当該強判別器よりも判別精度の低い前記強判別器の判別結果を入力し、複数の前記推定値及び入力された判別結果に基づいて前記判定画像に前記対象物体が描写されているか否かを判別すること、
を特徴とする物体検出装置。 - 前記強判別器は、複数の前記強判別器のうち当該強判別器よりも次に判別精度の低い前記強判別器の判別結果を入力する請求項1に記載の物体検出装置。
- 前記強判別器は、前記弱判別器の識別容易性の度合いを示す重みと当該弱判別器の推定値とを用いた重み付け投票の結果、及び入力された判別結果に基づいて前記判定画像に前記対象物体が描写されているか否かを判別する請求項1又は2に記載の物体検出装置。
- 前記強判別器は、当該強判別器よりも判別精度の低い前記強判別器の判別結果として、前記弱判別器の識別容易性の度合いを示す重みと当該弱判別器の推定値とを用いた重み付け投票の結果を入力する請求項1~3の何れか一項に記載の物体検出装置。
- 複数の前記強判別器には、判別精度に応じた重みがそれぞれ付与されており、
前記強判別器は、入力された重み付け投票の結果に入力元の前記強判別器の重みを積算し、積算された値を用いて前記判定画像に前記対象物体が描写されているか否かを判別する請求項4に記載の物体検出装置。 - 対象物体の特徴量に基づいて判定画像に対象物体が描写されている可能性を示す推定値をそれぞれ算出する複数の弱判別器と、複数の前記推定値に基づいて前記判定画像に前記対象物体が描写されているか否かをそれぞれ異なる判別精度で判別する複数の強判別器とを備える物体検出装置の物体検出方法であって、
複数の前記強判別器が判別精度の低い順に直列結合されて、判別精度の低い順に前記強判別器が判別を実行する実行ステップと、
前記強判別器が、当該強判別器よりも判別精度の低い前記強判別器の判別結果を入力し、複数の前記推定値及び入力された判別結果に基づいて前記判定画像に前記対象物体が描写されているか否かを判別する判別ステップと、
を含み、
前記判別ステップにおいて、前記強判別器が前記判定画像に前記対象物体が描写されていると判別した場合には前記実行ステップを継続し、前記強判別器が前記判定画像に前記対象物体が描写されていないと判別した場合には前記実行ステップを中断すること、
を特徴とする物体検出方法。 - 前記判別ステップでは、前記強判別器が、複数の前記強判別器のうち当該強判別器よりも次に判別精度の低い前記強判別器の判別結果を入力する請求項6に記載の物体検出方法。
- 前記判別ステップでは、前記強判別器が、前記弱判別器の識別容易性の度合いを示す重みと当該弱判別器の推定値とを用いた重み付け投票の結果、及び入力された判別結果に基づいて前記判定画像に前記対象物体が描写されているか否かを判別する請求項6又は7に記載の物体検出方法。
- 前記判別ステップでは、前記強判別器が、当該強判別器よりも判別精度の低い前記強判別器の判別結果として、前記弱判別器の識別容易性の度合いを示す重みと当該弱判別器の推定値とを用いた重み付け投票の結果を入力する請求項6~8の何れか一項に記載の物体検出方法。
- 複数の前記強判別器には、判別精度に応じた重みが付与されており、
前記判別ステップでは、前記強判別器は、入力された重み付け投票の結果に入力元の前記強判別器の重みを積算し、積算された値を用いて前記判定画像に前記対象物体が描写されているか否かを判別する請求項9に記載の物体検出方法。
Priority Applications (6)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2010513549A JP4806101B2 (ja) | 2010-02-01 | 2010-02-01 | 物体検出装置及び物体検出方法 |
| PCT/JP2010/051355 WO2011092865A1 (ja) | 2010-02-01 | 2010-02-01 | 物体検出装置及び物体検出方法 |
| US12/742,869 US8693791B2 (en) | 2010-02-01 | 2010-02-01 | Object detection apparatus and object detection method |
| EP10714836A EP2397989A1 (en) | 2010-02-01 | 2010-02-01 | Object detection device and object detection method |
| CN2010800008446A CN102216958A (zh) | 2010-02-01 | 2010-02-01 | 物体检测装置以及物体检测方法 |
| US14/179,041 US20140161364A1 (en) | 2010-02-01 | 2014-02-12 | Object detection apparatus and object detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/JP2010/051355 WO2011092865A1 (ja) | 2010-02-01 | 2010-02-01 | 物体検出装置及び物体検出方法 |
Related Child Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US12/742,869 A-371-Of-International US8693791B2 (en) | 2010-02-01 | 2010-02-01 | Object detection apparatus and object detection method |
| US14/179,041 Continuation US20140161364A1 (en) | 2010-02-01 | 2014-02-12 | Object detection apparatus and object detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2011092865A1 true WO2011092865A1 (ja) | 2011-08-04 |
Family
ID=44318872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2010/051355 WO2011092865A1 (ja) | 2010-02-01 | 2010-02-01 | 物体検出装置及び物体検出方法 |
Country Status (5)
| Country | Link |
|---|---|
| US (2) | US8693791B2 (ja) |
| EP (1) | EP2397989A1 (ja) |
| JP (1) | JP4806101B2 (ja) |
| CN (1) | CN102216958A (ja) |
| WO (1) | WO2011092865A1 (ja) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014530424A (ja) * | 2011-09-27 | 2014-11-17 | インテル・コーポレーション | Adaboost学習アルゴリズムを使用して顔特徴点の位置を検出するための方法、装置、およびコンピュータ可読記録媒体 |
| JP2015187759A (ja) * | 2014-03-26 | 2015-10-29 | キヤノン株式会社 | 画像検索装置、画像検索方法 |
| CN112686209A (zh) * | 2021-01-25 | 2021-04-20 | 深圳市艾为智能有限公司 | 基于车轮识别的车后盲区监测方法 |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012068948A (ja) * | 2010-09-24 | 2012-04-05 | Renesas Electronics Corp | 顔属性推定装置およびその方法 |
| US9449259B1 (en) * | 2012-07-25 | 2016-09-20 | Hrl Laboratories, Llc | Opportunistic cascade and cascade training, evaluation, and execution for vision-based object detection |
| CN103577789B (zh) * | 2012-07-26 | 2018-02-13 | 中兴通讯股份有限公司 | 检测方法和装置 |
| US9317768B2 (en) * | 2012-09-25 | 2016-04-19 | Intel Corporation | Techniques for improved feature detection |
| JP6111796B2 (ja) * | 2013-03-28 | 2017-04-12 | ブラザー工業株式会社 | 画像処理装置およびコンピュータプログラム |
| KR20150005094A (ko) * | 2013-07-04 | 2015-01-14 | 삼성전자주식회사 | 전자 디바이스 및 전자 디바이스에서 눈 영역 검출 방법 |
| WO2018119684A1 (zh) * | 2016-12-27 | 2018-07-05 | 深圳前海达闼云端智能科技有限公司 | 一种图像识别系统及图像识别方法 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7099510B2 (en) | 2000-11-29 | 2006-08-29 | Hewlett-Packard Development Company, L.P. | Method and system for object detection in digital images |
| JP2006268825A (ja) * | 2005-02-28 | 2006-10-05 | Toshiba Corp | オブジェクト検出装置、学習装置、オブジェクト検出システム、方法、およびプログラム |
| WO2008026414A1 (fr) * | 2006-08-31 | 2008-03-06 | Osaka Prefecture University Public Corporation | Procédé de reconnaissance d'image, dispositif de reconnaissance d'image et programme de reconnaissance d'image |
| JP2009087209A (ja) * | 2007-10-02 | 2009-04-23 | Sony Corp | 画像処理装置、撮像装置、これらにおける処理方法およびプログラム |
| JP2009116401A (ja) * | 2007-11-01 | 2009-05-28 | Canon Inc | 画像処理装置および画像処理方法 |
| JP2009151395A (ja) * | 2007-12-18 | 2009-07-09 | Canon Inc | パターン識別器生成方法、情報処理装置、コンピュータプログラム及び記憶媒体 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7769228B2 (en) * | 2004-05-10 | 2010-08-03 | Siemens Corporation | Method for combining boosted classifiers for efficient multi-class object detection |
| JP4767595B2 (ja) | 2005-06-15 | 2011-09-07 | パナソニック株式会社 | 対象物検出装置及びその学習装置 |
| KR100813167B1 (ko) * | 2006-06-09 | 2008-03-17 | 삼성전자주식회사 | 고속 정확한 얼굴 검출 방법과 얼굴 검출 학습 방법 및이를 구현하는 시스템 |
| JP4980988B2 (ja) | 2008-06-13 | 2012-07-18 | トヨタ自動車株式会社 | 運転者状態推定装置 |
-
2010
- 2010-02-01 EP EP10714836A patent/EP2397989A1/en not_active Withdrawn
- 2010-02-01 US US12/742,869 patent/US8693791B2/en not_active Expired - Fee Related
- 2010-02-01 JP JP2010513549A patent/JP4806101B2/ja not_active Expired - Fee Related
- 2010-02-01 WO PCT/JP2010/051355 patent/WO2011092865A1/ja active Application Filing
- 2010-02-01 CN CN2010800008446A patent/CN102216958A/zh active Pending
-
2014
- 2014-02-12 US US14/179,041 patent/US20140161364A1/en not_active Abandoned
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7099510B2 (en) | 2000-11-29 | 2006-08-29 | Hewlett-Packard Development Company, L.P. | Method and system for object detection in digital images |
| JP2006268825A (ja) * | 2005-02-28 | 2006-10-05 | Toshiba Corp | オブジェクト検出装置、学習装置、オブジェクト検出システム、方法、およびプログラム |
| WO2008026414A1 (fr) * | 2006-08-31 | 2008-03-06 | Osaka Prefecture University Public Corporation | Procédé de reconnaissance d'image, dispositif de reconnaissance d'image et programme de reconnaissance d'image |
| JP2009087209A (ja) * | 2007-10-02 | 2009-04-23 | Sony Corp | 画像処理装置、撮像装置、これらにおける処理方法およびプログラム |
| JP2009116401A (ja) * | 2007-11-01 | 2009-05-28 | Canon Inc | 画像処理装置および画像処理方法 |
| JP2009151395A (ja) * | 2007-12-18 | 2009-07-09 | Canon Inc | パターン識別器生成方法、情報処理装置、コンピュータプログラム及び記憶媒体 |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014530424A (ja) * | 2011-09-27 | 2014-11-17 | インテル・コーポレーション | Adaboost学習アルゴリズムを使用して顔特徴点の位置を検出するための方法、装置、およびコンピュータ可読記録媒体 |
| US9202109B2 (en) | 2011-09-27 | 2015-12-01 | Intel Corporation | Method, apparatus and computer readable recording medium for detecting a location of a face feature point using an Adaboost learning algorithm |
| JP2016001492A (ja) * | 2011-09-27 | 2016-01-07 | インテル・コーポレーション | Adaboost学習アルゴリズムを使用して顔特徴点の位置を検出するための方法、装置、およびコンピュータ可読記録媒体 |
| US9563821B2 (en) | 2011-09-27 | 2017-02-07 | Intel Corporation | Method, apparatus and computer readable recording medium for detecting a location of a face feature point using an Adaboost learning algorithm |
| JP2015187759A (ja) * | 2014-03-26 | 2015-10-29 | キヤノン株式会社 | 画像検索装置、画像検索方法 |
| CN112686209A (zh) * | 2021-01-25 | 2021-04-20 | 深圳市艾为智能有限公司 | 基于车轮识别的车后盲区监测方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| US8693791B2 (en) | 2014-04-08 |
| US20140161364A1 (en) | 2014-06-12 |
| CN102216958A (zh) | 2011-10-12 |
| JPWO2011092865A1 (ja) | 2013-05-30 |
| JP4806101B2 (ja) | 2011-11-02 |
| US20120020514A1 (en) | 2012-01-26 |
| EP2397989A1 (en) | 2011-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP4806101B2 (ja) | 物体検出装置及び物体検出方法 | |
| JP4657934B2 (ja) | 顔検出方法および装置並びにプログラム | |
| US7881494B2 (en) | Characteristic point detection of target object included in an image | |
| JP4712563B2 (ja) | 顔検出方法および装置並びにプログラム | |
| JP4933186B2 (ja) | 画像処理装置、画像処理方法、プログラム及び記憶媒体 | |
| JP4517633B2 (ja) | 対象物検出装置及び方法 | |
| US9202109B2 (en) | Method, apparatus and computer readable recording medium for detecting a location of a face feature point using an Adaboost learning algorithm | |
| JP4553044B2 (ja) | 集団学習装置及び方法 | |
| US20080175447A1 (en) | Face view determining apparatus and method, and face detection apparatus and method employing the same | |
| WO2014074959A1 (en) | Real-time face detection using pixel pairs | |
| JP6852779B2 (ja) | 画像認識装置、画像認識方法、および、画像認識プログラム | |
| JP4795864B2 (ja) | 特徴点検出装置および方法並びにプログラム | |
| WO2012046426A1 (ja) | 物体検出装置、物体検出方法および物体検出プログラム | |
| JP4624889B2 (ja) | 顔検出方法および装置並びにプログラム | |
| JP4588575B2 (ja) | デジタル画像の複数対象物検出方法および装置並びにプログラム | |
| JP2007094633A (ja) | 顔検出装置ならびにプログラム | |
| US7643674B2 (en) | Classification methods, classifier determination methods, classifiers, classifier determination devices, and articles of manufacture | |
| JP4749884B2 (ja) | 顔判別装置の学習方法、顔判別方法および装置並びにプログラム | |
| Hassan et al. | Facial image detection based on the Viola-Jones algorithm for gender recognition | |
| JP2014010633A (ja) | 画像認識装置、画像認識方法、及び画像認識プログラム | |
| JP7509212B2 (ja) | 情報処理システム、情報処理方法、及びコンピュータプログラム | |
| JP2011170890A (ja) | 顔検出方法および装置並びにプログラム | |
| JP2007108990A (ja) | 顔検出方法および装置並びにプログラム | |
| JP2007323104A (ja) | 目の状態判別方法および装置ならびにプログラム | |
| JP4795737B2 (ja) | 顔検出方法および装置並びにプログラム |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| WWE | Wipo information: entry into national phase |
Ref document number: 201080000844.6 Country of ref document: CN |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2010513549 Country of ref document: JP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2010714836 Country of ref document: EP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 12742869 Country of ref document: US |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10714836 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |