Research On Intrusion Detection Of Foreign Objects On Track Based On Image Processing

In the past,when intrusion detection was carried out on the foreign objects in the surveillance video on both sides of the railway and the road,manual active monitoring was mostly used,and it was difficult to promptly and effectively remind when encountering emergencies.With the development of network systems,if a real-time detection algorithm for foreign body intrusion in surveillance video can be developed,the rate at which abnormal conditions are discovered and handled can be greatly improved.In response to these problems,this article improves the methods related to the detection and tracking of moving targets in railway surveillance videos:(1)Because the Vibe algorithm has great advantages in the background modeling of video images,Vibe is used as the subject detection algorithm for orbital foreign objects in this paper.However,Vibe is prone to smear in the background modeling process of video images.To solve this problem,an adaptive Vibe detection algorithm based on improved LaBGen-P is proposed.The difference between this algorithm and LaBGen algorithm is that it does not require any training period to reduce the complexity of the algorithm.When establishing the amount of motion detection window,it takes into account the locality of pixels in the entire space domain to remove the original algorithm.When calculating the amount of motion,it is established by pixel iteration the subset improves the accuracy of the exercise amount calculation.After applying LaBGen-P to the initial stage of the background model of the Vibe algorithm,it can effectively solve the problem of smear in the background modeling process of the Vibe algorithm.At the same time,for the problems that are prone to occur in the foreground segmentation stage,this paper uses the method of adjusting the adaptive threshold in the spatial domain to improve.(2)For some algorithms,when performing target detection,the shadows of moving targets(track-side foreign objects)will be detected together as the output of the foreground target.This paper uses a color estimation model based on Horprasert for shadow removal.Calculate the expected value and actual color value of the pixel on the RGB color channel by the probability density estimation method,and then decompose it into two parts of brightness distortion and color distortion,and compare them with the set threshold respectively to determine whether it belongs to shadow.(3)For some algorithms,when tracking moving targets(foreign objects on the track)in track surveillance video,there may be situations where the target is blocked,so that the target is easy to lose.This paper proposes a fusion algorithm based on CamShift and Vibe to randomly extract the samples in the image,which improves the robustness of the target tracking algorithm.