Study On Road Pedestrian Recognition System Based On Adaboost Algorithm

With the rapid increase of car ownership in China, road traffic accidents often occur, especially vehicular and pedestrian collisions leading to accidents is the main reason leading to pedestrian casualties. This makes the application of the safe driver assistance systems have the greater urgency and practical significance in our country. Vision-aided navigation is one of the focuses of the vehicle safety driver assistance field, road pedestrian detection is an important part of the visual navigation system. In this paper, the main research is road pedestrian recognition based on image.Base-vision pedestrian detection system generally consists of two modules:the interested district segmentation and the object recognition, Monocular vision sensor is used as the means to access the information of the external environment. from a practical point starting, Establish a movement pedestrians segmentation and recogni-tion detection system.(1) Pedestrian segmentation is based on pedestrian shape information for image segmentation. First based on road recognition system access the pedestrian interest region, through vertical edge enhancement and edge threshold segmentation for the pedestrian interest region calculate the symmetry measure of the pedestrian, in order to determine the pedestrian symmetry axis and pedestrian width, determine pedestrian start edge by horizontal edges. Finally, based on the pedestrian aspect ratio to determine the pedestrian level and to achieve the pedestrian precise positioning.(2) Pedestrian recognition is through establishing a cascade AdaBoost classifier for pedestrian recognition. Pedestrian recognition phase consists of two parts:offline training and online recognition, in the offline training phase, according to the pedestrian shown the vertical edges and symmetry and used of Haar features describing pedestrian to get cascade AdaBoost classifier training; in the online recognition phase use the cascade classifier of the training phase obtained to recognize and detect the input pedestrian candidate region.The pedestrian detection algorithm was tested on the large number test set including the different weather and light conditions, Test results show that the proposed method can identify effectively of different sizes, colors and shapes of the pedestrian who in front of the vehicles, for static and dynamic pedestrian have good test results, the real-time system is better.