Research Of Moving Object Tracking And Recognition Based On Video Surveillance

Intelligent monitoring video as a new generation of security means.Use the computer instead of the artificial work to detect,analyze,track and identify the video sequence.With the implementation of the "smart city",digital video information is increased.Artificial alone is unable to complete the work of information analysis,So the intelligent surveillance video technology research is necessary.Computer vision technology involves a lot of domain-related knowledge,such as image processing,pattern recognition,artificial intelligence.At present,there are many achievements in the field of research,but the monitoring of the application of the complex environment,and can not simply copy the results of various fields.Target tracking and identification system requires good real-time and stability,to ensure that the system can be in a variety of complex scenes can be completed quickly and accurately.Therefore,the robust and real-time algorithm is the key problem of target tracking and recognition algorithm.In this paper,according to the needs of surveillance video target tracking and identification technology research,the main work is as follows:In this paper,a three-frame difference target detection algorithm with background difference is proposed to analyze the algorithm of various motion target detection.Based on the characteristics of frame difference method and good adaptability,this algorithm proposes a block background extraction method based on frame difference method to improve the complexity and real-time difference of mixed Gaussian model background extraction.For the frame difference method,it is easy to use the background difference method to set the threshold to judge the performance of the frame difference method when dealing with objects with similar color.If the frame difference method is poor,the background difference method is used.Experiments show that this method has improved the detection rate and real-time compared with the traditional algorithm.Aiming at the requirement of monitoring video,a histogram of LBP texture and chroma is proposed,which can suppress the lighting problem and occlusion problem in monitoring video.The size of the object in the surveillance video will vary from far to near,and the viewing angle will change in part,using the Camshift algorithm to solve these problems.Analysis of domestic and foreign identification technology,the HOG feature is applied to the target object recognition algorithm based on feature learning.Collecting data for weak classifiers to be trained and cascaded weak classifiers.The algorithm is used to test the test data and analyze the experimental results.In summary,this paper proposes a new target detection algorithm based on the requirement of monitoring video,and completes the better target tracking and recognition algorithm,which can be applied to various monitoring environments such as city monitoring and intelligent transportation.