Research On Pedestrian Detection Method For Unmanned Vehicles

Pedestrian detection and tracking technology based on machine vision applications in the field of intelligent vehicles become more and more popular.It can help the driver to understand the surrounding environment better,identify the surrounding pedestrians,reduce traffic accidents and damage to pedestrians.This paper studies the detection and identification of monocular machine vision on road pedestrian ahead.The main work of this paper is divided into three parts,Firstly,the input image is needed to be preprocessed.By the weighted average gray scale algorithm,median filter algorithm denoising,Sobel operator edge contour detection and moment invariant method binarization operation can reduce the computational complexity and improve the detection speed.Then,the common pedestrian characterization operator is analyzed.HOG descriptors are geometrically and optically invariant,so HOG descriptors are especially suitable for human detection.In this paper,HOG features are selected as pedestrian characteristics to be extracted,and positive and negative samples are trained by support vector machine.By changing the number and proportion of positive and negative samples to achieve a good classification effect.On this basis,the new classifier is trained with LBP and the second training,reducing the false detection rate to improve the effect of the classifier.By the experiments which video and pictures collected in the campus the algorithm has good robustness and accuracy.And also to meet the needs of real-time.And the algorithm is ported to other processor devices,to facilitate the development.