| Pedestrian detection and distance estimation method based on machine vision,which is an important part of the intelligent driving assistance system,can provide important information such as pedestrian location and distance between the pedestrian and the vehical accurately for the driver and effective guarantee for traffic safety.In this paper,the pedestrian detection algorithm based on the deformable part-based model(Deformable Part Model,DPM)and the distance estimation monocular algorithm based on monocular vision are concerned.The specific work in this paper includes the following aspects:1.In this paper,the pedestrian detection theory is introduced in detail from three aspects: pedestrian feature extraction and representation,pedestrian detection classifier and the evaluation standard on pedestrian detection algorithm.The common pedestrian detection datasets are presented.2.The improved HOG feature extraction,feature pyramid construction,non-maximum suppression and DPM model structure in DPM are studied in detail.The HOG+SVM detection algorithm and DPM detection algorithm were tested on the ETH data set.The test results showed that the DPM detection algorithm had better detection accuracy.Because the DPM detection algorithm adopts multi-component model for joint detection,the detection efficiency is not high.In view of this deficiency,this paper proposes a method of approximate estimation of adjacent scale features to construct the feature pyramid in DPM detection algorithm.The improved algorithm in the ETH test data sets,the experimental results show that,compared with the DPM detection algorithm,on the premise of guarantee the accuracy,this algorithm improves the detection efficiency of detection algorithm,the average per frame detection time is 0.16 s.3.On the basis of obtaining pedestrian position,this paper USES monocular vision distance measurement method to obtain vehicle and human distance.Firstly,the image coordinate system and its transformation and the calibration of zhang’s camera are described in detail.Then the pedestrian and camera distance formula is obtained through the geometric relationship between the small hole imaging model,the imaging point and the pedestrian target.The experimental results show that within 10 m,the average error of the algorithm in this paper is within 0.2m,and the range measurement model in this paper can accurately calculate the distance between people and vehicles,so as to achieve the goal of distance measurement. |
