Research On Motion Pedestrian Detection Algorithm And FPGA Implementation In Video Surveillance

Intelligent video surveillance systems have the advantages of intelligence and digitalization,which are widely used in transportation and security now.With the introduction of management systems such as smart city and safe city,the application of intelligent video surveillance is more extensive.As an important part of intelligent video surveillance,motion pedestrian detection has attracted more and more attention.In this paper,the motion foreground detection and motion pedestrian detection algorithms are mainly studied.The main work is as follows:Firstly,the traditional background subtraction method based on pixel modeling is optimized,and the adaptive background subtraction method based on pixel block is applied to motion detection.In the foreground detection,the background model is initialized by the minimum difference sum algorithm and which is selectively updated by the information entropy estimation algorithm.The threshold calculated by the maximum inter-class variance method is applied to the background subtraction method to obtain the binary image of the motion foreground detection.The MATLAB simulation is carried out and compared with other algorithms.The results show that the motion prediction algorithm based on pixel block has better detection effect.Then,for the binary result of the detection,a run-length-based one-scan connected domain labeling algorithm is used to extract the feature information and the MATLAB simulation proves which can detect the motion pedestrian better.In order to improve the real-time performance of motion pedestrian detection,this paper builds an FPGA-based hardware platform,including image cache module,algorithm processing module and image output module.The algorithm processing module includes motion foreground detecting unit and motion pedestrian detecting unit.In the motion foreground detection unit,the Verilog hardware circuits about adaptive threshold calculation,background initialization and background update which based on pixel block are designed.In the motion pedestrian detection unit,the hardware circuits of run-length connected domain label and feature information extraction are designed.And Modelsim simulation analysis are performed on each module,and then board level verification is completed.Finally,the hardware results are compared with the software results with an error of 0.08%,meeting the design requirements.The experimental results in this paper show that the background subtraction method based on pixel block can detect the binary image of motion foreground more accurately;the run-length-based one-scan connected domain labeling and extraction algorithm can detect motion pedestrians better;and the motion pedestrian detection system based FPGA can achieve a processing speed of 29 fps with a resolution of 640*480 images,thus that enables real-time detection.