Optimization And Implementation Of Real-time Pedestrian Detection Based On C66x-DSP

Pedestrian detection has always been the research focus and hotspot in the field of computer vision,it is an important basic technology applied in many occasions.It has a wide range of applications in intelligent video surveillance,virtual reality(VR),and robotics.However,most of the pedestrian detection algorithms are designed and implemented on the PC platform,which has weak engineering practicability.With the rapid development of image processing technology,the real-time performance of video sequence processing tasks on most hardware platforms still cannot meet the needs..TMS320C66x series of embedded DSP processors of TI company have powerful signal processing ability and high-speed parallel processing ability.They have unique advantages in real-time video processing and is an excellent platform for hardware implementation of pedestrian detection algorithm.This project selected TI's TMS320C6657 chip and implemented real-time pedestrian detection in the video sequence on the corresponding DSP development board.This design mainly researches the pedestrian detection technology in the video with the camera fixed.In terms of detection algorithms,in view of the disadvantages of slow recognition and low efficiency when using sliding window detection in the traditional method,according to the relevant application scenarios,the application of pedestrians is always in motion,a novel pedestrian detection model by combining a background model based on vibe and histogram of oriented gradient(HOG)plus support vector machine(SVM)is proposed.Firstly,extract motion regions in video by vibe algorithm to get the m foreground image,and the forground image is corroded,expanded and connected to extract the final area of interest.Then,the pedestrian detection algorithm based on HOG feature and SVM statistical classification is used to detect the extracted moving target region,and mark the pedestrian target.In the optimization and implementation of the algorithm on the DSP,first of all,according to the programming characteristics of DSP,we arrange and rewrite the pedestrian detection algorithm on the PC platform to implement the embedded DSP platform transplantation of the algorithm.Then,using the unique optimization technologies of DSP platform,including system-level compiler optimizations,cache optimizations,and module-level software pipelines,inline function optimizations,and VLIB libraries,etc.while ensuring the detection accuracy,fully mining optimization resources to improve the detection speed.The experimental results show that the pedestrian detection system can detect pedestrians more accurately on the embedded DSP platform.The optimized monitoring system can realize real-time detection of pedestrians with a resolution of 768 × 576 video at 24f/s,which has certain engineering practicality value.