TWI420429B - Night traffic detection method combined with fuzzy theory - Google Patents

Description

Night traffic detection method combined with fuzzy theory

The invention relates to a night traffic detection method for combining the fuzzy theory which can accurately detect the traffic flow in the night environment, and is particularly suitable for the road segment or the like where each vehicle traffic is to be detected.

The image-based traffic detection method of the conventional image is separated from the vehicle image by the establishment of the background model, such as the "Import-type Traffic Parameter Automatic Detection System" of the Patent No. I298857 and the "Traffic Monitoring System" of the I293746 Patent, the above-mentioned The method can accurately detect the traffic flow in a clear daytime environment, but it cannot work normally due to brightness problems such as insufficient lighting in the night environment.

For the nighttime environment, most vehicles are used to detect vehicles. For example, China Patent Publication No. I302879 "Intelligent Night Vehicle Instant Detection and Identification System Based on Computer Vision" is based on image recognition in the car unit. Measuring nighttime front lights; if applied to intersection units, such as references 1 to 3, the position of the lights is positioned by the paired characteristics of the lights; wherein the paired characteristics of the lights are through coordinates, area, density , and so on, a number of parameters are determined together; therefore, it is not easy to determine an optimal threshold value, thus resulting in an unsatisfactory detection effect. It can be seen that there are still many shortcomings in the above-mentioned conventional methods, which is not a good design and needs to be improved.

In view of the shortcomings derived from the above-mentioned conventional methods, the inventor of the present invention has improved and innovated, and after years of painstaking research, he finally succeeded in researching and developing the night traffic detection method based on the fuzzy theory.

The main object of the present invention is to provide a nighttime traffic detection method combining the fuzzy theory which can accurately detect the traffic flow in a nighttime environment with poor brightness or illumination.

A secondary object of the present invention is to provide a nighttime traffic detection method that combines fuzzy theory with a low cost and easy installation.

Another object of the present invention is to provide a night traffic detection method that can be combined with other digital monitoring functions and integrated into a smart monitoring system.

In order to achieve the above object, the present invention mainly comprises the following steps: a. binarizing the bright object image with a dynamic threshold; b. detecting and locating the position of the bright object on the binarized image; c. tracking the bright object Moving the track; and d. identifying the characteristics of the bright object; thereby, the instant input image is transmitted through the brightness analysis, and the fuzzy theory is used to identify the pair of lights; finally, the moving track of the running light in the continuous image is analyzed, and the calculation is performed. Corresponding parameters such as steam and locomotive flow are used to enhance the ability to resist light and shadow in the night environment, which helps to calculate more precise traffic parameters in the night environment.

According to the above steps, first, the bright object binarization processing is performed on the input image, and the bright objects in the screen are separated from the image, and an optimal binarization threshold is automatically determined according to the screen content, and then the bright object is detected. Position the plane coordinates at the same time, and then analyze the continuous binarized image to track the moving trajectory of the bright object. Finally, by identifying the characteristics of the bright object, including analyzing the dynamic and static features of the object, determine whether it is a car. Pairs of lights, single lights of the locomotive or unknown light and shadow interference, and then obtain the corresponding traffic flow information.

According to the nighttime traffic detection method combined with the fuzzy theory, the binarized dynamic threshold is an iterative method, and the number of iteration operations is determined by the group separation degree and the lowest group number threshold.

According to the nighttime traffic detection method combined with the fuzzy theory, the position of the detected and positioned bright object is obtained by the connected calibration method to obtain the number of objects and their coordinates.

According to the nighttime traffic detection method combined with the fuzzy theory as described above, the tracking trajectory of the bright object is transmitted by the distance method as a basis for tracking.

According to the nighttime traffic detection method combined with the fuzzy theory as described above, the tracking is based on the input of multiple image features as a fuzzy inference engine, and defines the degree of object similarity as the output of the fuzzy inference engine.

In the nighttime traffic detection method combined with the fuzzy theory as described above, the input system of the fuzzy inference engine includes defining a membership function, blurring the obtained image features, and using the maximum and minimum operators as inference tools of the fuzzy inference engine.

The nighttime traffic detection method combined with the fuzzy theory as described above, the output of the fuzzy inference engine includes the use of the centroid method as a tool for defuzzification.

The nighttime traffic detection method combined with the fuzzy theory provided by the invention has the following advantages when compared with other conventional technologies:

1. Special research and development for the night traffic monitoring environment, especially the high-brightness contrast generated by the lights in the night environment as a feature of detection, which can reinforce the problem of insufficient lighting in the night environment, thereby improving the recognition rate in the night environment. .

2. The method of binarization of bright objects used is to determine the optimal grouping by using the two characteristics of separation degree and the lowest grouping number. Compared with the conventional technology, it can effectively bright objects in the complex environment of light and shadow. Separation.

3. The method of tracking bright objects used is to match and track objects by distance. It can analyze all objects at the same time and track multiple objects to facilitate the analysis of trajectories for bright object recognition.

Referring to FIG. 1 , the nighttime traffic detection method combining the fuzzy theory of the present invention mainly includes the following steps:

a. Binding of bright object images with dynamic threshold values;

b. Detecting and locating the location of the bright object on the binarized image;

c. Tracking the movement of bright objects;

d. Identify the characteristics of bright objects.

Please refer to FIG. 2 for the detection flow chart of the present invention. After the image recognition process 2 receives the image 1 captured by the intersection monitor, a series of image recognition and analysis procedures are performed. The steps are as follows: Step 20: Dynamic Threshold Value Perform bright object image binarization; Step 21 to detect and locate the location of the bright object; Step 22 to track the movement of the bright object; Step 23 Identify the characteristics of the bright object, the pair of lights enter Step 24, and the single light enters Step 25, The headlights proceed to step 26; step 24 accumulates the car counter; step 25 accumulates the locomotive counter; step 26 has no action; step 3 outputs flow information.

Please refer to Appendix I to Annex V. Attachment 1 is the original image taken by the intersection camera. The original image is input into the image recognition process 2 of the present invention, and the binary image of Annex 2 is generated. The binarization of the bright object is mainly The gray-scale value statistical graph is analyzed for the gray-scale components of the color image; and according to the number distribution of the gray-scale values, the optimal clustering is determined by the degree of separation and the lowest clustering number, and the group with the highest gray-scale value is determined. As the basis for the binarization threshold.

The image processed by the bright object is binarized, and then the bright object is detected and positioned. The bright object detection uses the Connected Component Labeling to obtain the rectangular region of the object and its corresponding coordinate point; when the rectangular region satisfies a minimum The area threshold is treated as a reasonably bright object, and then the trajectory of the bright object is tracked. As shown in Annex III, the tracking method is to record the moving trajectory and its trajectory characteristics by analyzing the reasonably bright objects in the continuous input image. Wherein, the bright object tracking includes a method corresponding to the detecting object, and the method is responsible for one-to-one correspondence between the detecting object of the current input image and the detecting object of the input image at the previous moment, and the corresponding basis is accordingly The distance between the different detected objects is matched, and the two short-distance two-detecting objects are matched to the same tracking object; since the image is continuously captured, the displacement difference of the same tracking object in the front and rear images is extremely small, and is not easily adjacent. Objects interfere with, and Annex 4 is a schematic diagram of the trajectory of detecting bright objects according to the present invention.

Step 23 identifies that the bright object characteristics can be used to identify whether the tracked object is a pair of lights, a single headlight, or an unknown light and shadow interference. When all of the tracking objects begin to characterize 24, the fuzzy theory is used to determine whether the two objects are paired 25, and then the unmatched tracking objects are divided into a single light 26 and a light interference 27.

The determination of the pair of headlights is based on five characteristics of "tracked number of sheets", "level of overlap", "area similarity", "degree of parallelism" and "object distance", and each feature can pass the experiment. Each group determines an optimal set of critical values. The set of critical values will be combined with a variance to form a membership function of the fuzzy inference. When the two objects enter the comparison program, the membership function determines whether the two objects match or not. A degree value that satisfies all features; the degree value is fed into a fuzzy inference engine defined by the present invention, which will test each item with the "max-min operator" with reference to the individual degree values of the five features. The fuzzy inference rule, in which the fuzzy inference rule defines each possible input combination and its corresponding output result; then, the result of each inference law is deduced from the "center of gravity" to the degree of similarity between the two objects. The possible outcomes are four possibilities: "paired", "should be paired", "may be paired" or "unpaired". After any two tracking objects pass this procedure, the degree of similarity of the two objects can be obtained; finally, all the objects are determined according to their similarity or not. If there are remaining tracking objects that are not matched, the unknown light and shadow interference of the vehicle will be filtered according to its travel path and trajectory characteristics (step 26), and the unfiltered tracking object will be regarded as a single vehicle light of the locomotive. Finally, according to the paired characteristics of the lights, the vehicle flow and the locomotive flow in the monitoring environment are determined; the fourth is the original image of the intersection containing the steam and the locomotive, and the fifth is the steam and locomotive after the analysis of the present invention. One embodiment can be tracked separately.

When the bright object in the picture analyzes the pair of headlights and the single headlight through the fuzzy inference, when the trace of the headlight crosses the preset counting line in the picture, it is accumulated to the car counter according to the pairwise or not (step 24) or the locomotive counter (step 25), and count the lanes according to the position of the lights to achieve the traffic flow statistics (step 3); the recorded data can be provided by day, hour or unit Report statistics and remittance functions to achieve effective traffic information control.

The detailed description of the preferred embodiments of the present invention is intended to be limited to the scope of the invention, and is not intended to limit the scope of the invention. The scope of the patent in this case.

In summary, the present invention is not only innovative in terms of technical ideas, but also provides a nighttime environment with poor brightness or illumination, which can accurately detect traffic flow, and has low device cost, easy installation, and can be matched with other The structure with digital monitoring function is integrated into a nighttime traffic detection method combining fuzzy theory with a smart monitoring system. Therefore, the present invention fully complies with the statutory invention requirements of novelty and progressiveness, and submits an application according to law.

a. . . Bright object image binarization with dynamic threshold

b. . . Detect and locate the location of bright objects on the binarized image

c. . . Track bright object movements

d. . . Identify the characteristics of bright objects

1. . . Input image

2. . . Image recognition process

20. . . Bright object image binarization with dynamic threshold

twenty one. . . Detect and locate the location of bright objects

twenty two. . . Track bright object movements

twenty three. . . Identify the characteristics of bright objects

twenty four. . . Identify the car lights and accumulate the car counter

25. . . Identifying the locomotive lights and accumulating the locomotive counter

26. . . Recognize light and shadow interference, do not count

3. . . Output flow information

Figure 1 is a flow chart of the main structure of the present invention.

Figure 2 is a block flow diagram of the present invention.

Annex I: Night road image map.

Annex 2: The binarized image of step 20 of the present invention.

Annex III: Step 22 of the present invention tracks the movement of bright objects.

Annex IV: Night road image 2

Annex 5: Step 22 of the present invention tracks the movement trace 2 of the bright object.

a. . . Bright object image binarization with dynamic threshold

b. . . Detect and locate the location of bright objects on the binarized image

c. . . Track bright object movements

d. . . Identify the characteristics of bright objects

Claims (7)

A nighttime traffic detection method combining fuzzy theory includes the following steps: a. binarizing a bright object image with a dynamic threshold; b. detecting and locating a bright object at a binarized image; c. Tracks the movement of bright objects; and d. identifies the characteristics of bright objects. For example, in the nighttime traffic detection method combined with the fuzzy theory described in claim 1, the binarized dynamic critical value is an iterative method, and the iteration is determined by the group separation degree and the lowest group number threshold. The number of operations. For example, in the nighttime traffic detection method combined with the fuzzy theory described in claim 1, the position of detecting and locating the bright object is obtained by the connected calibration method to obtain the number of objects and their coordinates. For example, in the nighttime traffic detection method combined with the fuzzy theory described in the first application of the patent scope, the tracking trajectory of the bright object is tracked by the distance method. For example, in the nighttime traffic detection method combined with the fuzzy theory described in claim 1, the plurality of image features are used as input of the fuzzy inference engine, and the degree of object similarity is defined as the output of the fuzzy inference engine. For example, the nighttime traffic detection method combined with the fuzzy theory described in claim 5, wherein the input of the fuzzy inference engine includes defining a membership function, blurring the obtained image features, and using the maximum and minimum operators as blurring Inference engine inference tool. For example, the nighttime traffic detection method combined with the fuzzy theory described in claim 5, wherein the output of the fuzzy inference engine includes the use of the centroid method as a tool for defuzzification.