Research And Implementation Of Pedestrian And Two-wheel Vehicle Detection Based On Vision

Great convenience has been brought to human by rapid development of science and technology, which gives people’s liberation from the heavy labor. Special thanks to the vehicles, there will be no longer myth on the daily long distance traveling, and the quality of going out has been incredibly improved. However, at the same time of promoting the convenience of peoples’ trip, the vehicles also brings a lot of traffic problems. The main reason is the imbalance between the increasing speed of the road construction and the rising speed of the vehicle number. More and more traffic accidents have happened increasingly because of the gradually crowding of the traffic. Pedestrians and vehicles which are the main members of the road, their safeness has become the social problem which can not afford to ignore. Concerning about the safeties of the pedestrians, the vehicles and the drivers, it is important to avoid the accidents by reminding the drivers when dangerous is about to come.The paper has focused on researching and accomplishing the detection and tracking algorithm of pedestrian and two-wheel vehicle based on monocular vision. In the reality, there are many kinds of vehicles, including bicycles, motor vans, the electric vehicles, limousines, large vans, etc. Basing on the research of the pedestrian detection, the paper mainly makes a deep research on two-wheel vehicles. The characters of two-wheel vehicle pedestrian include both the characters of walking pedestrian and the complicated vehicle, and the speed of two-wheel vehicle is faster. So it is also significant for researching two-wheel vehicle pedestrian. The paper mainly carries on researches and experiments on motor vans, because of their most complicated constructions, fastest speed, and the easiest occurrence of the traffic accidents in all of two-wheel vehicles.The paper mainly researches and accomplishes the following areas:First, do a series of pretreatment on the gathered videos, including the vision transforms, edge detections, noise eliminations, etc. Secondly, get the cascade classifier by training on a number of positive and negative samples. Then, using the classifier to identify the target, and improve the accuracy of detection by putting in the detections and judgments on the wheels. Finally, track and forecast the target which is beyond the detection. The paper adopts the combination of the wheel detection and the modified AdaBoost algorithm to identify the target, as well as the combination of CamShift and Kalman filtering to track and forecast the target.The experimental results show that:the methods the paper uses have the abilities to identify and track the pedestrians and the two-wheel vehicles. Both the real-time of the system and the accuracy of the identification have been improved. The robustness of the algorithm under different environments is especially good.