Design Of Intelligent Front-end System For Personnel Video Detection

With the widespread use of intelligent monitoring systems,a large number of distributed network cameras upload video data,which causes channel congestion and puts enormous computational pressure on the server.In order to achieve a certain amount of calculation forward and reduce the computing pressure of the server,this paper designs a set of intelligent front-end system for personnel video detection.The system can extract the skeleton features of the moving human body in the video and achieve a certain workload forward.The main work done in this paper includes:(1)From angles such as algorithm complexity,resource utilization,analyzes the advantages and disadvantages of several common motion target detection method,and puts forward a kind of fusion on the basis of frame differential method to improve the prospects for a moving target and Vi Be algorithm,solves the common motion target detection algorithm can’t complete motion target prospects.(2)In view of the two defects of ZS thinning algorithm in extracting skeleton results,that is,the skeleton has redundant information and cannot guarantee the integrity of the skeleton,a refinement processing method of the refined image is proposed to ensure the single-pixel width of the skeleton,remove redundant redundant information,and pass the ZS algorithm on the basis of the deletion conditions.Increase the retention conditions to avoid the occurrence of fractures.Finally,the MIT data set is used for experimental verification.Compared with the ZS refinement algorithm,the algorithm in this paper solves the problem of fracture of the extracted target skeleton,and reduces the redundant information by 12% on the basis of ensuring the integrity and connectivity of the skeleton.After obtaining the single pixel wide skeleton,the skeleton is modeled by obtaining the relative position information of key joint points,and the amount of information to be transmitted is reduced by this method.(3)Through the modular design idea,the system is divided into each specific function module,and then the hardware realization of each module is completed with the thought of pipelined program design.Finally,the input and output of each module are connected through the relationship between each function module to complete the hardware realization of the system as a whole.On the system under the environment of the actual experiment many times,using a skeleton precision evaluation index to analyze the experimental results,in this paper,the designed system can effectively extract the motion human body in the video frame characteristics and obtain the key joint point location information,the method of using transmission key key points,greatly reduces the need to transfer the amount of data,The system has the ability to process continuous video sequences in real time with the maximum processing speed of 98 ms per picture.