Research On The Moving Target Classification Based On Multi-feature In Intelligent Video Surveillance System

At the same socio-economic development, and human activities also will be able tochange the scope of broad surveillance video data generated also rose to massive levels.This allows the video image data processing and analysis, and to extract usefulinformation intelligent video surveillance system to be more widely used. In this context,how can detect moving targets accurately classified in accordance with the expectedclassification, has been in intelligent video surveillance system was particularly prominentin public places such as security surveillance, intelligent traffic monitoring, etc. have beengreatly concerned. Intelligent video surveillance including video image sequence movingtarget detection, tracking, classification and behavior analysis process. By comparing withthe study of existing moving target detection method chosen as the research-basedbackground modeling method based on a random sample. Moving target trackingalgorithm to learn a simple nearest neighbor tracking, and on top of this method isproposed based on linear prediction nearest neighbor tracking.Moving target classification technology can not only reduce the workload ofmonitoring personnel, but also provide the basis for subsequent analysis and processing ofmoving targets. Further work on the basis of the former, the first goal of the currentcommon characteristics were analyzed. And the SVM learning and classification methodsK-means, and on this basis propose an online semi-supervised classification method basedon SVM and K-means is. This algorithm can classify the unknown moving target based onthe existing classification, and can target online learning.Since moving targets tend to be more sensitive to color surveillance personnel, thusmoving target color classification has become important in intelligent video surveillanceadjunct. In this paper, moving object classification method based on HSV color space.Algorithm is verified by experiments in this paper has a high accuracy and robustness,whether simple or complex scenes in the scene are able to meet real-time requirements.