Research On Traffic Police’s Command Posture Recognition Based On OpenCV

In recent years,with the growing of driverless technology and increasing of intelligent automobile and people’s travel frequency,traffic congestion,accidents and signal lights in trouble occur from time to time.At this time,it is necessary for traffic police to conduct on-site command to clear the road and maintain traffic order.At present,there are few researches on traffic police command action recognition based on video and most of them have some limitations.Therefore,this paper puts forward a method to recognize the traffic police command posture action based on video sequence and with the help of OpenCV visual library.The specific research content of this article includes:(1)Methods for detecting traffic police in video sequences.In order to solve the problem that most of the traditional traffic police detection methods have a large amount of calculation and the detection accuracy is greatly influenced by the noise of sample and the complexity of background,and combining with the characteristics of the traffic police’s own clothing,a method of traffic police detection based on color gamut screening is proposed Firstly,use R,G,B three color channels to filter out the object of non-target color.Then set a certain threshold of the dark channel to select the image of traffic police.(2)Traffic police gesture sequence preprocessing and video action segmentation.Firstly,the traffic police command video is sampled in time domain;then each frame is preprocessed to obtain the normalized sequence of noise suppression;finally,the whole traffic police command video is processed by the method of the extreme value of motion trajectory.And then obtaining a sub-video containing only one kind of traffic police gesture signal,which reduces certain calculation of action feature extraction and recognition.(3)feature extraction of dynamic body contours of traffic police.The traditional detection algorithm is vulnerable to environment and illumination,and there are many noise points and incomplete contours in the extraction results.So the paper puts forward a method of dynamic region edge point retention.However,there are fracture points and unclear parts at the edge of the contour extracted by this method,so the method of integrating the dynamic edge points is proposed for filling.Finally,the experiments showed the improved method can obtain more accurate edge contours from video sequences.(4)Traffic police command posture action classification recognition.According to the characteristics of small video sample and few kinds of gesture action,the DTW template matching method is selected to classify and identify the action.Experiments show that the recognition time and recognition rate of DTW algorithm are almost the same.It is found that the recognition time of the two methods is almost the same,which verifies that the greater the action complexity,the longer the recognition time of the two methods,the lower the recognition rate of method one,the weaker the correlation between the recognition rate and the action complexity of method two;the average recognition rate of method one is higher,reaching 89.50%,and the method two is only 84.13%.All of these have some guiding significance in the process of using DTW algorithm in the future.With the help of OpenCV4.0.0、Visual Stidio 2017,all experiments in this paper are realized by C++programming,and the method is verified by using the traffic police command posture action set established by myself,and the real-time and accuracy of the improved method are verified by comparing with the traditional method.