Research On Pedestrians Rapid Detection Based On Machine Vision

In recent years, road traffic accidents occur frequently in our country. The harmto our social is self-evident. Therefore, how to minimize the frequent traffic accidentsand reducing its loss has become the focus of attention among people. In our country,a large proportion of victims of road traffic accidents victims are pedestrians andcyclists. Therefore, to carry out road traffic pedestrian protection is imperative. Ourgovernment and many related research institutions have carried out a lot of researchwork. Pedestrian detection technology will provide technical support to the intelligentvehicle safety driver assistance technologies. Pedestrian detection technology hasgreat potential application value. In this background, this paper carry out a pedestriandetection method for pedestrian safety, aimed to protecting pedestrians and cyclists.In the road transport system, due to the diverse of body’s clothes, body’s posture,body’s scale and the different light and background of the environment, undoubtedly,pedestrian detection technology facing great challenges. In2005, Dalal et al. putforward human detection method based on the histogram of oriented gradient (HOG).the method is simple in principle. This method achieved nearly90%detection rate Inthe case of10-4FPPW (False Positive Per Window). Today, the HOG method is stillone of the excellent methods in the field of human detection. INRIAPerson sampleshas also become a recognized standards people detection samples. Because of this, theresearch of pedestrians rapid detection facing to the safety of pedestrians is based onthe HOG algorithm.Although the HOG algorithm is very good, it is difficult to achieve the real-timepeople detection in the road system. itself is imperative to optimize HOG algorithm.The optimization of HOG algorithm is as follows:(1) This study found that using the Integral histogram of oriented gradient(IHOG) method can greatly speed up the process of feature extraction. The Integralhistogram of oriented gradient method only need to scan the entire image for oncetime and storage the Integral gradient data, can any area’s HOG feature be obtainedwith simple addition and subtraction operations, without repeat to calculate theorientation and gradient of each pixel in the area. (2) In order to avoid time-consuming in the pedestrian training and detectioncaused by too many HOG feature vector dimensions (3780). This paper found theregions of interest (ROI) of people by looking for the samples’ significant map. TheROI contains four regions: the pedestrian head, the left arm, the right arm. Theinterest regions of people have1764dimensions feature vector. This paper putforward the ROI-IHOG method using integral histogram in the ROI-HOG featurevector.(3) In order to adapt the road system when detect people, this paper establishedperson database of road system (RSPerson). This paper use SVM to trainingpedestrian classifier base on INRIAPerson and RSPerson.In order to verify the ROI-IHOG method and the SVM classifier’s performance,this paper doing offline test and online test. The result shows that the detection speedof ROI-IHOG method4times faster more than the original HOG method. Althoughthe dimensions of people features reduced more than one half, we obtain a betterdetection result. The ROI-IHOG method can achieve real-time pedestrian detection inthe320pixels×240pixels video and with the rate of scale change at1.2. Thepedestrian classifier is good at finding pedestrian in the road environment.