The Research On Pedestrian Detection Method For Intelligent Driving

The realization and popularization of advanced intelligent assistant driving system and even automatic driving will greatly improve the current situation of traffic congestion and effectively reduce frequent traffic accidents.Pedestrian detection technology as the core of automatic driving technology,its accuracy,real-time,complexity will directly determine the development of automatic driving.Pedestrian detection technology based on machine vision has been widely studied and applied in automatic driving,intelligent monitoring,intelligent robots and other fields because of its simple and intuitive framework and strong adaptability.However,due to the non-rigid characteristics of pedestrians and the influence of various costumes,illumination,occlusion and other complex factors,it is very difficult to accurately detect pedestrians.Based on the analysis of the principles and advantages and disadvantages of various pedestrian detection technologies,this paper mainly studies the enhancement of pedestrian characteristics and the reduction of the impact of local pedestrian information on the pedestrian as a whole.A pedestrian detection method based on the fusion of HOG and Gabor features and the idea of deformable component model is proposed.The main contents and conclusions of the two studies are as follows:(1)In order to compensate for the deficiency of describing pedestrian information with single feature,this paper combines HOG feature which has good descriptive ability with Gabor feature which has good descriptive ability with pedestrian edge and texture information to form new enhanced features for pedestrian detection by weighted fusion,and does it on the larger INRIA data set and the more complex self-made data set SIC at present.The comparative experiment of detection was carried out.(2)In order to reduce the false alarm rate and the false alarm rate of pedestrians under occlusion,a fast pedestrian recognition method based on the fusion of features and the idea of deformable component model and the feature pyramid estimation is proposed,which combines the head,shoulder and leg of pedestrians,and the performance test is carried out.The experimental results on INRIA and SIC datasets show that the average detection rate of HOG + Gabor fusion feature method is 6.9% higher than that of singlefeature method on two datasets;the detection rate of INRIA and SIC datasets using two-part combination method based on fusion feature is 91.5% and 88.3% respectively,and the detection rate is higher than that of HOG + Gabor fusion feature method.5.6%and 7.1% prove the validity of this method.