Research And Implementation Of Human Abnormal Behavior Recognition Based On Hisilicon Platform

The current research on intelligent video surveillance is mostly based on computers,and the video data is transmitted to the computer for processing through the monitoring terminal.The disadvantage of this is that the more numbers of monitoring terminals,the greater pressure of data processing on the computer.To solve above problem,the research content of this article is that the video sequence analysis is completed at the monitoring terminal based on Hisilicon Hi3531 embedded chip,and the result of analysis is transmitted to the server.The recognition of human behavior is mainly divided into three steps:the extraction of the human motion foreground(including the relevant pre-processing),the characteristics extraction of human behavior and the classifier training and recognition of human behavior.In terms of foreground extraction,this paper analyzes the principle,advantages,disadvantages and differences of temporal difference method and background difference method in detail.Through experimental comparison of the effects of the two extraction algorithms,the two-frame difference method is selected as the foreground extraction algorithm of this paper.The foreground pre-processing introduces morphological filtering.The opening operation and closing operation can be deduced from the erosion operation and dilatation operation of binary images.In regard to characteristics extraction,the extracted foreground is accumulated to obtain the Motion Energy Image(MEI).The Hu moment descriptor and the Fourier characteristics descriptor are extracted on the MEI the key frame.And the translation,rotation,scaling and invariance of starting point of the fourier descriptor are analyzed.In the classification and recognition,Naive Bayesian classifier based on the Bayesian Theory is trained by the use of the sample video,and the specific behavior of human falling is identified by the training result.In this paper,the hardware processor is the embedded chip Hisilicon Hi3531.Firstly,the hardware parameters and interface of the chip Hi3531 and the media process platform(MPP)are introduced in detail.Then the necessity and steps of the establishment of cross-compilation and network file system are detailedly described in Hisilicon Linux development environment.Finally,this paper presents the design plan of intelligent monitoring system based on Hisilicon platform,which consists of video capture device,main processing chip Hi3531,monitoring server and video display device.Among them,the main processing chip Hi3531 completes the identification of the human fall behavior and the transmission of the recognition result to the monitoring server.At the same time,the monitoring server can receive the real-time monitoring video data compressed into format H.246.Hu invariant moments are used as characteristics descriptors in the board level test.The results show that all falling behaviors can be detected and the alarming message can be timely transmitted in the real-time surveillance.