Design Of Monitoring Terminal With Video Analysis Function Based On HiSilicon 3518E

In recent years,the domestic video surveillance industry has developed rapidly under the impetus of large-scale networked monitoring projects such as Safe City,Wisdom,and Skynet.Relying on the new generation of video processing chips,the functions of video surveillance terminals are constantly improving and gradually applied to all aspects of people's lives.At present,the processing power of embedded platforms is getting stronger and stronger,and it can support the transplantation of many video analysis algorithms.Compared with the method of implementing the video analysis function by using the PC client,the video analysis function implemented on the embedded monitoring device can effectively reduce the pressure on the server side and save the storage cost.Based on this background,this paper conducts in-depth research on network video surveillance equipment based on embedded Linux,and designs and develops an embedded monitoring terminal with video analysis function.The specific research contents are summarized as follows:(1)By analyzing market user habits and actual project needs,Haisi Hi3518EV200 chip is selected as the core processor of the monitoring terminal.The chip has H.264 & JPEG multi-stream real-time encoding capability,high internal module integration,and can control hardware cost and terminal size.The hardware circuit and software function design is carried out around the main chip,and finally an intelligent monitoring terminal with high integration and various functions is completed.(2)Monitoring terminal hardware circuit design,mainly including chip minimum system design and peripheral function circuit design.In order to save user cost and reduce maintenance difficulty,this paper proposes an external CCD sensor for video capture,which avoids the phenomenon of discarding the entire terminal due to sensor damage.Moreover,the CCD image sensor wiring is more flexible,and the transmission data is not distorted,which can ensure better video collection quality.Considering the diversification of data transmission and increasing the fault tolerance of equipment operation,two kinds of data transmission methods,WIFI and MAC port,are designed.In order to meet the needs of users to control outdoor terminals in the room,a PTZ connection circuit was designed.(3)Monitoring terminal software function implementation.Firstly,based on the HiSilicon MPP framework,the video image acquisition and encoding functions are realized,and the BT656 format data input and H.264 encoding output are supported.In this process,in order to solve the problem of adaptation of video capture hardware design and MPP software functional framework,the underlying code such as video mask configuration and video buffer pool configuration has been greatly modified.Then the "client-server" communication framework based on RTSP protocol is built.Considering the compatibility of the monitoring terminal,the ONSOIF protocol framework based on gSOAP is built,and all network video client connection terminals that conform to the ONVIF protocol specification are supported and obtained.Video data.Finally,the RS485 serial port is debugged,and the PTZ control function based on the Pelco_d protocol is implemented.The driver and function configuration of the WIFI module are performed.(4)Design and implementation of video analysis functions.In the framework of background difference method,a new local background modeling(LBM)method and global foreground modeling(GFM)method are proposed for target detection.In this method,a new feature vector is introduced,which integrates RGB values,Haar wavelet features in horizontal and vertical directions,and time difference features of pixels,increasing the dimension of feature vectors to improve their discriminative power;For the problem of missed detection,the GFM method globally selects the Gaussian density to model the foreground pixels according to the Bayesian minimum error decision rule.When an object gradually stops moving,the foreground model will still maintain accurate Gaussian density to model;Considering that the amount of data that face recognition needs to be processed in the future will become larger and larger,this paper uses the Eigenface method to implement face recognition.By using the monitoring terminal as a distributed processing node,parallel processing of analysis tasks is realized,thereby improving people.The speed of face recognition.