TW201006527A - Measuring object contour method and measuring object contour apparatus - Google Patents

Abstract

A measuring object contour method is provided. The measuring object contour method includes the following steps: capturing multiple sequence images including a current image and at least a previous image; deriving a motion image from calculating the motion difference of the sequence images; deriving an edge map from operating with the current image; deriving a motion edge map from operating with the motion image; superposing the edge map with the motion edge map to derive a foreground edge map from extending the contour in the edge map and corresponding to the motion edge map. Besides, a measuring object contour apparatus using the measuring object contour method mentioned above is also provided.

Description

201006527 IX. Description of the Invention: [Technical Field of the Invention] The present invention relates to a method for analyzing an object contour, and in particular to a method for removing an object contour moving under the background (4) (4), to realize an interface for human-computer interaction. . [Prior Art] 技术 The technology of 互动 互动 介 has been greatly developed and developed recently, especially in the game industry, where players hope that the game itself can directly feedback the player's actions and enjoy the game. In the human-computer interaction interface, most of the current practice is still that the user wears special clothing or wears a special logo marker. The receiver can sense the user's movement, but this is quite a child's benefit when using 2 words. . Therefore, it is a very intuitive way to use the visual type as a human-machine. Material, the current image mutual = only (4) measuring all the images in the area of change or activity and the environmental background or identification efficiency and other factors, it is not easy to grasp complete and accurate • = contour 'is therefore not effective for the user's limb movements A detailed method for determining the B/f ° flag to sense the user's action has been disclosed in U.S. Patent No. 6,308,565. In the United States, the 5524637 'Y is mainly to paste the active or passive marker on the user 2, 'by detecting the amount of movement of the marker to calculate the patent speed of the user's force, and then achieve the interface of human-computer interaction. . In the United States No. 6308565, it is still necessary to attach the active or passive logo to the user-specific 'Fin by the detection target' of the amount of movement' with coordinate mapping to the positioning side 201006527

The method obtains the 2D/3D space coordinate position of the user's movement, and then interacts with the virtual reality. As mentioned above, since these methods of identification need to be worn on the user, it is very inconvenient for the user. In addition, it is disclosed in Japanese Patent No. 6,307,951 that the action of the user is determined purely by image processing. It is mainly based on blocks, and analyzes the brightness change of each block, and further serves as a reference for calculating the momentum change of moving objects. Finally, the Fourier transform is used to calculate the moving force and direction of the moving object area. However, the second right user only waved his arm while the rest of the body remained motionless. Then the above method can only extract the user's arm (moving the object contour ' and regard the part that the user has not moved as the background, but cannot get the whole body, the contour. So - this method is interpreting the user action The above-mentioned deviations can be made so that it is impossible to achieve the human-computer interaction interface and the fine-grained interaction only by image processing. [Invention] In view of this, the object of the present invention is to provide an object for analyzing the contour of an object. The outline can be used to analyze the part of the user's movement, and the extension of $ extends to the overall outline of the user, thereby achieving the interface state of human-computer interaction. ^ The profile of the user or other purposes, the present invention proposes a Analyze the object wheel k package. The following steps: capturing a plurality of sequence images, and the sequence I: the eye-catching image and the at least one previous image; performing operations on the sequence images, and "motion image"; For the current image ^ different 'to get the edge map (edge map); to calculate the momentum map, 6 201006527 edge map (motion edge map); and stack eight edge map:: map 'and self-corresponding FIG edge of the edge:: Aberdeen the edge view of FIG eg gamma ship nd edge map). In the embodiment of the wheel variation, the pixel values of the pixels corresponding to the image corresponding to the sequence image are subtracted from the previous image c pixel values to obtain a momentum map. ~

In an embodiment of the invention, the foregoing step A 计算 ' 计算 计算 计算 计算 计算 计算 计算 计算 计算 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 这些 动 动 动 动 动 动 动To obtain a plurality of difference maps; and to add pixels to obtain a momentum map. In an embodiment of the present invention, the above-mentioned center-to-center: in: step may employ a spatial edge filtering method. In an embodiment of the present invention, the step of moving the momentum edge map The image binarization method may be used to perform the steps of: setting _; the momentum graph greater than _ includes the following, and the pixel value of the small (four) value of the numerator is set in the present invention - the actual _ t The upper pre-image is set to 0. = like information, and the object image information includes coordinates, squares: the object foreground edge map further includes the action rule data: the image information of the object of the judgment, wherein the action rule database has a button comparison table and the semantic comparison table includes the former Move, move backward, right shift and left shift. In the actual example of this month, the above steps of uncovering the sequence image are

201006527 Capture, machine or network camera with image class. The image capturing unit may be digitally photographed. In an embodiment of the present invention, after the image information, the object image information of the foreground edge image may be transmitted to the display unit by using the upper layer, and the interaction may be performed. Object TV or projection TV. The unit may be a plasma television or a liquid crystal. In the eighth aspect of the present invention, the edge map and the momentum edge map are obtained. In the method for analyzing the contour of the object, the contour is a well profile, and the momentum edge map includes only the edge map of the object. The light edge map including all the backgrounds is superimposed on the momentum edge and the partial rim is moved. By i = _ ground 1, the profile is used to automatically exclude the pure background contour, and the overall outline of the object. After that, you can make a «° analysis of the outline of the object, especially if it is under the user's object, you can reach the interface state of each other. The above and other objects, features, and advantages of the present invention will become more apparent. [Embodiment] FIG. 1 is a schematic diagram of a method for analyzing an object contour according to an embodiment of the present invention, and FIG. 2Α~2D are images obtained by the method for analyzing a contour of an object according to the present invention, respectively. Figure. Please refer to FIG. 1 first, as in step S11 ώί·-, ^. The invention firstly captures a plurality of sequence images, and the sequence of the shirt images includes the current image and the previous image. In this embodiment, the image capturing unit continuously captures the sequence images of the object in front of the 201006527 撷 image capturing unit, and the sequence images have images of the background object and the foreground object, wherein the foreground object can be regarded as The object that the user will move, and the background object is a table, a chair, etc.; an object that does not move. Further, the image capturing unit may be a digital camera or a web camera or the like, but the invention does not limit the type of image capturing unit. In the above, the sequence images captured may be array image images, that is, each sequence image has a plurality of arrayed pixels, and each pixel has a specific pixel value to indicate color, brightness, and the like. . These sequence images are sorted by time series, and the latest captured sequence image is called the current image, and the previously captured sequence image is called the previous image. Then, as shown in step S12, the momentum changes are calculated for the sequence images to obtain a momentum map, and the spirit of the momentum map is the "degree of change" of the sequence image. In this embodiment, the momentum map is obtained by subtracting the pixel value of the pixel of the current image from the pixel value of the corresponding pixel of the previous image. In general, the previous image here selects the sequence image whose time series is closest to the current image, that is, if the current image is time series η, the previous * image will select the sequence image with time series η-1. . - In this case, in the momentum graph, a pixel whose pixel value is equal to zero indicates that the region is in a stationary state, and a pixel whose pixel value is not equal to zero indicates that the state of the region has changed. This allows you to distinguish the "variation" of the sequence image. It should be noted that the foregoing calculation method is merely an example of how to form a momentum map by the concept of "degree of change", but the present invention does not limit the calculation method of the momentum map. For example, in the current image with time series η, 9 201006527 can be selected first to take the previous images of time series n-l and n-2. Calculate the difference between the current image of the time series n and the previous image of the time sequence of n-1, and calculate the difference between the current image and the time series of the time series as the previous image. The quantity map, the two difference maps are summed to obtain the momentum map. Those skilled in the art will be able to modify the calculation of the momentum map slightly according to the foregoing description, but it is still within the scope of the present invention. Please refer to Figure 1, even if it is a very sophisticated image capture unit, there will still be interference from sway, noise or stray light. That is, in the case where the external objects are not moved, the pixel values of all the pixels of the momentum map are still not completely zero. Therefore, as shown in step S13, momentum measurement is then performed on the momentum map to determine whether the momentum map indicates object movement, or only noise. If the _ momentum map does indicate that there is an object moving, then the subsequent steps are entered; if the Hungarian momentum map is only noise, then the process returns to step S11 and continues to capture the sequence image. Incidentally, the present invention may also skip the step after the determination of step S13 and the subsequent steps, but if there is a judgment flow of step S13, there is a better analysis effect. In the present embodiment, the momentum calculation is performed by the slanting momentum map in a statistical manner. In detail, in this embodiment, the gate rate "4" is first set in the momentum map. If the ratio of the number of pixels whose pixel value is not zero exceeds the threshold rate, then the momentum map is determined to indicate the movement of the object, and vice versa. The picture shows the noise. When the threshold is 1/16' and the resolution of the momentum map is 1024x728 pixels, then when there are 46592 (1024x728/16) pixels whose pixel value is not zero, then the momentum map is not considered as noise. Carry out the next steps. Then, as shown in steps S14 and S15, the current image is computed to obtain the edge map '201006527 and the momentum map is calculated to obtain the momentum edge map', wherein the edge map is as shown in FIG. 2A' and the momentum edge map is as Figure 2b ^8" It is stated at a glance that steps S14 and S15 have no order, and it is convenient for length = 曰°, and step S14 is performed first, and then step S15 is performed. However, it is easy for the skilled person to understand Step S15 is performed first, then step " or step S14, S15 is performed simultaneously.

❹ In step S14 of the embodiment, the current image is calculated by using the spatial edge filtering method or the spectral edge filtering method to 彳β map (Fig. 2Α), wherein the Sobel operator method is a spatial edge two-wave ^ Kind, and can have better results. In short, the edge map is the “sketch” of the current image. The image in the current image is discontinuous and the continuous and discontinuous portions of the image are binarized (1, 〇), and the edge image in black and white is obtained (Fig. 2Α). In this embodiment, = the user of the foreground and the stationary chair of the background are included, so white = that is, the outline of the user and the chair, and the area that does not belong to the wheel will be indicated by the black dot. Similarly, in step S15 of the present embodiment, the momentum map is directly calculated by the binarization method to obtain a momentum edge map (Fig. 2b), i.e., the momentum edge map is the "sketch" of the momentum map. In the present embodiment, the image binarization method first sets a threshold value, and the threshold value is mainly used for interference such as interference, and the threshold value can be set according to the actual pixel value (4). In order to take the above, all the pixels of the momentum map are divided into two categories, that is, the pixels whose pixel values are larger than the threshold are classified into -classes, and the pixels whose pixel values are smaller than the threshold are classified into another class - and respectively have two values (1 G) means that the momentum edge map in black and white can be obtained (Fig. 2B). In the present embodiment, since only 11 hands are moving, the white point of the momentum map corresponds to the two-handed instant. The rim, while the other unmoved parts and the background are in the black point table: refer to Figure 2A. The foreground part (white point area) in the edge map is the rim of the mover's rim'. Part (white point area) is a static chair

The movement, in which the user's arm portion is moving, causes the edge map (white point region) of Figure 2B to reflect the contour of the moving arm portion. The edge ^ is further referred to FIG. 1 and as shown in step S16, the overlapping edge image and the momentum scene edge are taught to extend from the contour of the edge image of the corresponding momentum edge image, and the front view and the momentum are shown in FIG. The edge map is superimposed as shown in Figure 2c and the % edge map is shown in Figure 2D. In detail, move (the pixel value is the pixel of !, the white part of the picture is like the seed"'s seed of any momentum edge map (in terms of outlines, there will be a seed corresponding to this seed at the same position on the edge map). 1 and as shown in Figure 2Γ pixels, ^ pixels (indicated by black dots).

1 pixel (black dot) is found in the vicinity of the same contour M. In the picture, it is represented by the arrow, and then continuously extends outward. After all the pixels of the edge map corresponding to the seed are extended and expanded (the figure is only extended by two pixels), the edge of the scene can be expanded from the edge map (Fig. 2D). In this embodiment, the seed of the momentum edge map; = should be in the user's hand, so that the user's hand contour can be extended outward to form the user's overall contour. It is worth noting that the difference between the edge map and the foreground edge map is that the mineral edge map has no background outline of the chair, leaving only the foreground =. This is because the "seed" of the momentum edge map is extended from the contour of the hand to the contour of the user as a user's hand 12 201006527. In contrast, since there is no corresponding "seed" in the chair area, the outline of the chair does not appear in the foreground edge map (Fig. 2D). In this way, the present invention succeeds in moving the hand profile from the user, thereby expanding the overall contour of the user, and automatically removing the stationary portion 2 to obtain a foreground edge map. In this (4), the foreground edge map is two-dimensional, and the information can correspond to the user's action without the noise interference of the background object (chair). Therefore, by using the plurality of phase front edge L to calculate the object image of the user's coordinates, direction or speed, the image information of these objects can be used to further analyze the effect of the two-way interaction. Please use the 则 则 则 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后 之后Specifically, the 'action rule database' has semantic meanings to use the user's actions in response to changes in the outline of the object in the foreground edge map.

Current status Table 1 Object contour movement

201006527 Move right to the right by more than 80 pixels. Right shift does not move. 5 images move less than 10 pixels. Right shift to the right. Move more than 10 pixels. Move left. Move more than 80 pixels to the left. Move forward. No motion. 5 image movements. Less than 10 pixels move forward to move more than 10 pixels, then move up more than 25 pixels, move backwards, continue to move 5 images, move less than 10 pixels, move up, move more than 10 pixels, move forward, move more than 25 Pixel Table 1 lists a partial semantic comparison table. Please refer to Table 1. When analyzing the contour of the object in the image (as shown in Figure 2D), the meaning of the action used can be analyzed through Table 1. Of course, the semantic comparison table of this embodiment only performs four kinds of actions: up, down, left, and right. Those who are familiar with the art can easily expand the content of the semantic table and list more action semantics. Corresponding conversion relationship with the image outline. ❹ In general, the present embodiment can transmit the object image information of the foreground edge image to the display unit output by analyzing the action semantics of the object image of the foreground edge image. In this way, the user in front of the display unit can instantly see that his or her own action is displayed by the display unit, thereby achieving the effect of human-computer interaction or virtual reality. Further, the display unit is, for example, a plasma TV, a liquid crystal television or a projection type television, but the present invention is not limited to the type of display unit. In view of the foregoing concept of the method of analyzing an object contour of the present invention, the present invention further discloses an apparatus for analyzing an object contour, and the apparatus for analyzing an object contour is mainly applied to the aforementioned method. In order to further clear the device, the following will be combined with the illustration of the monthly analysis of the object wheel _ 圏 3 for (four) this (four) (four) _ block diagram. Referring to FIG. 3, the side of the contour device includes an image capturing unit 110 and an object rim device 1 〇 0 110 for processing the sequence image 112, = m' and the image capturing unit computer module U0. And wherein the sequence images m to the image information are included therein, and the user 5 can include the user's own position (for example, the simulation of the netting: the activity of the Dinger is continuously moved into the output unit 122 , the sequence image, etc. „). The computer module 120 includes a rim unit m, wherein the input wheel and the analysis object unit 11G and the sequence image memory unit 2 2 4 are coupled to the image removal unit m and coupled to the sequence image The memory profile analyzes the contour of the object to receive the above-mentioned 'rounding rim S - „ receiving sequence image 112, and the sequence is used to capture the sequence image 112 from the image capturing unit U, and :::==... is used to store the sequence image The foreground edge _ body wheel (four) element 126 is used for analysis. In detail, the analysis object wheel image 112 is analyzed to obtain the front edge 70 126 for the analysis sequence 叱~S16, that is, the sequence scene;艮: For the current image of step 51 of the above figure i, the operation of the current image ° (four) read Call the momentum soil row operation to correct the edge edge _ 2A), and the momentum map edge map (Fig. 2C), with the automatic and overlapping edge map and the momentum 2 bucket front with detailed map extension to obtain the foreground edge map (Figure Ocean These steps will not be repeated here. ",,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 1〇〇 is further connected to the analysis object rotation element, = and the translation unit m is a coupling library (not shown) to the present I-% 26, and utilizes the built-in action rule information. In addition, the display unit n Action ~ Translate unit 128, to display the 语~疋 lightly connected to the computer module 120, the semantics of the user 50 to interact. ^, the edge map of the object image information, and with the above, the road aa profile device at least There is a lower = excellent = analysis object rotation method and analysis object wheel includes; there is an edge map and a momentum edge map 'where the edge map is on the line borrowing = 脎 far, and the momentum edge map only includes the contour of the moving part of the object. #由的边图和动量边图 corresponds to the edge of the rim*## ^ Locate the outline of the edge and extend it as a reference extension. The background contour of the moving part can be moved from the object (the outline of the unwrapped body can automatically exclude the purely moving background), and the object can be accurately obtained. Wheel f^v ❹ - a two-way communication interface that is semantically separated by the contour of the object. 』Incoming each other, although the preferred embodiment of the present invention is as disclosed above, the reduction is not intended to limit the familiarity of the present invention. The skilled person will make some modifications and refinements without departing from the scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims. 1 is a flow chart 201006527*» of a method for analyzing an object contour according to an embodiment of the present invention. 2A to 2D are image views of the method for analyzing an object contour according to the present invention, respectively. Figure 3 is a block diagram of an apparatus for analyzing an object contour according to an embodiment of the present invention. [Main component symbol description] 50 : User

100: analysis object contour device 110: image capturing unit 112: sequence image 120: computer module 122: input and output unit 124: sequence image memory unit 126 · analysis object contour unit 128: semantic translation unit 130: display unit S11~ S17: Step 17

Claims (1)

201006527 X. Patent Application Range: 1. A method for analyzing an object contour, comprising: capturing a plurality of sequence images, wherein the sequence images comprise a current image; and at least one previous image; calculating momentum changes for the sequence images to Deriving a momentum map; a computing the current image to obtain an edge map; performing an operation on the momentum map to obtain a momentum edge map; and superimposing the edge map and the momentum edge map, and A contour edge map is obtained by extending the contour of the edge map corresponding to the momentum edge map. 2. The method for analyzing an object contour according to claim 1, wherein calculating the momentum change for the sequence image is subtracting a pixel value of a pixel of the current image from a pixel value of a corresponding pixel of the previous image, The momentum map is obtained. 3. The method for analyzing an object contour according to claim 1, wherein the number of previous images of Tan is two, and the step of calculating momentum changes for the sequence images comprises: 像素 respectively pixel values of pixels of the current image Subtracting the pixel values of the corresponding pixels of the previous images to obtain a plurality of difference maps; and • adding the difference maps to obtain the momentum map. 4. The method for analyzing an object contour according to claim 1, wherein the step of calculating the current image to obtain the edge map is a spatial edge filtering method or a spectral edge filtering method. 5. The method of analyzing an object contour according to claim 1, wherein the step of calculating the momentum map to obtain the momentum edge map is performed by image binarization. The method of analyzing an object contour according to claim 5, wherein the step of calculating the momentum map to obtain the momentum edge map comprises: setting a threshold; and using the momentum greater than the threshold The pixel value of the pixel of the figure is set to 1, and the pixel value of the pixel of the momentum map smaller than the threshold is set to zero. 7. The method for analyzing an object contour according to claim 1, wherein the foreground edge map has an object image information. 8. The method of analyzing an object contour according to claim 7, wherein the object image information comprises a coordinate, a direction or a speed. 9. The method for analyzing an object contour according to claim 7 is characterized in that after obtaining the foreground edge map, the object image information of the foreground edge map is determined by an action rule database. 10. The method of analyzing an object contour according to claim 9, wherein the action rule database has a semantic comparison table. ❹ 11. The method of analyzing an object contour as described in claim 10, wherein the semantic comparison table includes forward, backward, right and left shifts. 12. The method of analyzing an object contour according to claim 1, wherein the step of extracting the sequence images is performed by an image capturing unit. 13. The method of analyzing an object contour according to claim 12, wherein the image capturing unit is a digital camera or a webcam. 14. The method of analyzing an object contour according to claim 9, wherein after determining the image information of the object in the foreground edge map, the method further comprises: 19 201006527 s interacting with a user. 15. The method of analyzing an object contour according to claim 9, wherein after determining the image information of the object in the foreground edge map, the method further comprises: transmitting the image information of the object to a display unit. 16. The method of analyzing an object contour according to claim 15, wherein the display unit is a plasma television, a liquid crystal television or a projection television. 17. An apparatus for analyzing an object contour, comprising: an image capturing unit adapted to capture a plurality of sequence images; a computer module comprising: an input and output unit coupled to the image capturing unit to receive the image capturing unit a sequence of image memory units coupled to the input and output unit for storing the sequence images; and an analysis object contour unit coupled to the sequence image memory unit for analyzing the sequence images Prospect edge map. 18. The analysis object contour device according to claim 17, wherein the sequence images comprise a current image and at least one previous image, and the analysis object contour unit calculates momentum of the sequence image. Transforming to obtain a momentum map, calculating the current image to obtain an edge map, calculating the momentum map to obtain a momentum edge map, and superimposing the edge map and the momentum edge map to The foreground edge map is derived from the momentum edge map. 19. The analysis object contour device of claim 1, wherein the computer module further comprises a semantic translation unit coupled to the analysis object contour unit, and the semantic translation unit is adapted to use an action rule data. Library 20 201006527 'Determines object image information of one of the foreground edge maps. The analysis object contouring device of claim 19, further comprising a display unit coupled to the semantic translation unit of the computer module to display the object image information of the foreground edge map. Reference twenty one