Implementation And Optimization Of Pedestrian Detection Algoirthm Based On HOG+SVM In DM6437

Currently， the frequent occurrences of traffic accidents at night threaten people’s livesand property to a large degree.The main reason leading to collision and traffic accidentsbetween vehicles and pedestrains is the road confusion caused by the poor light.In infraredimage，wi can find targets at night or in poor light circumstance， and it doesn’t be interferedby glare and flash.Hence，pedestrian detection technology based on infrared image can realizepedestrian dectection at night or under hostile environment，which has important practicalutility in reducing the incidence of traffic accidents and ensuring pedestrian’s safety.In recent decades，video processing technologied have been rapid development，but it isdifficult to meet large volume of data and real-time requirements for the most hardwareplatforms.While TI’s Davinci series of DSP processors have a powerful parallel ability andplenty of resource，and have a special advantage in real-time proceesing and products’ rapiddevelopent，thus Davinci series of DSP processors are ideal choice for the implementation ofpedestrian detection.This paper mainly studies the pedestrian detection algorithm， and focuses on theimplementation and optimization of algorithm based on the DM6437.Specific work is asfollows:1、Through referring to a lot of domestic and international literatures，based-on infraredimage’s pedestrian detection algorithm’s development and research status are analysised， andthe TI-DSP processor’s special advantage in video processing is also explained.2、Introduced the basic working process of pedestrian detection system， and alsointroduced the foundational principle of HOG feature extracting algorithm and SVMclassifying algorithm， also introduced the implementation of the whole system procedureusing OpenCV2.1library.3、Introduced the implementation of HOG method on DM6437，TI，based on DSP IDECCS; alse take advantage of well-trained SVM liner classifier to predict the the pedestrianfrom candidate objects. 4、Using DSP/BIOS as the main structure of the whole pedestrian detection system， andusing the character of C64x+core and a variety of optimization strategies to optimize the DSPprogram. Experimental results show that this method performs well and has already satisfythe algorithm’s real-time real-time requirement.