Pedestrian Detection Based On Infrared Video

Nowadays, traffic accidents are important social problem. Most of the injured are the pedestrians in the traffic accidents.For this reason,pedestrian detection is the field of intelligent vehicle research frontier in recent years, and some auto manufacturers, universities and research institutions have started to study pedestrian detection technology. Because pedestrians are not rigid objects and their positions change everytime. Also there are many natural objects like people which bring the pedestrian detection many problems. This paper study algorithms of the pedestrian detection based on infrared image recognition.Because there is a lot of infrared image noise, first the infrared image will be preprocessed to filter the infrared image noise, and then based on the principle of thermal imaging of the infrared image,we use double binarization threshold method to detect image highlights. Last according to the infrared pedestrians characteristics of symmetry, and also the use of pedestrian aspect ratio,we split out about pedestrian area.The major factors of affecting the quality of the pedestrian classification performance are the samples, characteristics and classification algorithms. We use infrared camera shooting a lot of video in the selection of sample. We cut out a rich variety of positive samples and negative samples using hand-split approach which prepare a very good material for harr-like features extraction. Because visible video is more clearly than infrared video in texture and color, we use harr-like features which are more sensitive on the gray information in the pedestrian feature extraction. The location of pedestrian is generally brighter than oher areas in the infrared video which provides convenience for the Harr-like features extraction. In classifier selection,we use Adaboost classifier which is very fast calculation to classify pedestrians. First a number of weak classifiers are trained to form a strong classifier. And then the strong classifier to identify the pedestrian on the image. Adaboost classifier computes fast and reaches the goal of real-time detection of pedestrians. Finally this pedestrian detection algorithm test images of different scenes, the results show that the proposed algorithm for static and motion pedestrians has a good detection.