Research And Implementation Of Monocular Vision-based Pedestrian Detection Algorithm

Pedestrian detection is the field of intelligent vehicle research frontier in recent years, and some auto manufacturers, universities and research institutions have started studyed Pedestrian Detection Technology.Pedestrian detection and recognition algorithms have been introduced in the thesis, it mainly using pedestrians and legs of the vertical symmetry of features for segmentation in the pedestrian detection algorithm. When the direction of pedestrian movement and the camera at an angle an ego-motion compensation model is generated based on features matching in sequence frames to eliminate camera moving to generate difference image; then the difference image is used to detect motion.In the difference image using the edge of the pedestrian leg of the vertical symmetry and the characteristics of pedestrian segmentation.In the recognition process, we primarily use line support vector machine to identify pedestrians. we mainly through the selection and features of the sample to increase the recognition rate of pedestrians in the thesis.There are three main factors in the Classification performance on the pedestrian:features, the sample selection,classification algorithm.(1) characteristics of the HOG feature is the main advantage of the edge direction histogram, SIFT operator and shape of the deformation texture improvement comes. Its main idea is:the object of local appearance and shape can usually be through the local edge direction histogram or distribution of that came out fairly well. Fast HOG using the integral map features Traverse in the block by a certain percentage of the time traversing picture.(2) our sample of the calibration samples processed and used in the training of positive and negative samples when the ratio of1:2.(3) classification algorithm mainly use the linear SVM classifier and HOG features to quickly identify pedestrians.Pedestrian detection algorithm evaluation, the main video is the city of Shenyang City section of the road traffic around the scene, covering more comprehensive road conditions and various weather conditions, and make pedestrian detection algorithm test results show that the proposed algorithm for pedestrians (including static and motion) has a good detection.