Research On Target Recognition And Tracking Of Jitter Video Based On CNN

The automatic processing and prediction of surveillance video information has attracted great attention in many fields such as information science,computer vision,machine learning,pattern recognition and so on.In the process of acquiring surveillance video by monitoring system,complex and changeable environmental factors will have a certain impact on video acquisition.Video jitter is one of the adverse effects.Aiming at the target recognition and tracking in jittery video,because the location of the detected target in each frame of the video will change greatly,the traditional target detection method has no obvious effect on the target extraction in jittery video.Therefore,this paper proposes a deep learning convolution neural network(Faster R-CNN)model for target recognition in jittery video.Firstly,a large number of learning samples are trained to get classifiers by using convolutional neural network model.Then a series of candidate regions are selected by using Selective search algorithm.Then,the features of candidate regions are extracted by using convolutional neural network model.Then,the trained classifiers are used to classify and recognize targets.Then,the detected targets are fed back to the original dithering with regression model.In order to achieve target location and recognition in video,the improved Camshift algorithm is used to track the target by adding the features of the target that have been obtained.Compared with the original video detection model,the detection effect of jitter video is greatly improved in this paper.