Background Modeling Techniques To Target Perimeter Intelligent Video Surveillance

Moving target detection is one of the core technologies of video and image processing, which plays a key role in the high level video and image processing. The background modeling is a key technology of the moving target detection. Based on an overall study of three popular background modeling methods, this paper analyzes two traditional gaussian background modeling algorithms, i.e. the single gaussian background modeling algorithm and the gaussian mixture background modeling algorithm. At the same time, the paper discusses the PRBF neural network and the theory of the image orthogonal transformation and their application in image modeling.The paper develops a video and image feature extraction method based on the Walsh-hadamard transformation and introduces an adaptive learning update strategy to improve the performance of the traditional gaussian background modeling algorithm. The work brings two improved mixed gaussian background modeling methods to the world. One is the mixed gaussian background model based on PRBF neural network which exploits the adaptive learning ability of the artificial neural network; the other is based on Walsh-hadamard transformation which takes the advantage of Walsh-hadamard transformation’s ability to concentrate energy. Compared to the original mixed gaussian background model, the former helps to improve the moving object detection performance when the video and image bears a periodic dynamic changing background, while the latter can overcomes the problem of large background estimation error of the traditional gaussian mixture background model when the illumination condition changes considerably.A moving target detect module based on DSP is designed in this paper for real practical application. The module adopts our improved gaussian mixture background model algorithm with the Blackfin561DSP as the main processor. Based on the designed module an intelligent video monitoring system is developed. The system improves the performance of the traditional video monitoring system and is of strong practicability.