Research On Vehicle Front Mounted Pedestrian Detection Algorithm

Monocular vision-based pedestrian detection occupy an important position in the vehicle security system, which has become one of the most important research topics in monocular vision and vehicle safety.Its core is used a camera installed in vehicles up to detect pedestrians, in order to predict the potential danger that may occur and to take timely and appropriate measures to protect pedestrians.Pedestrian detection system target is to create a relatively independent, intelligent pedestrian detection system in a moving car, and this system can improve the security of drivers and protect pedestrians life and property security.With the progress of society, the related technology research developing, pedestrian detection technology has had some success in the environment of controlling. However, pedestrians are quite complex details of the changes which make the pedestrian detection more difficult.Mainly reflected in pedestrian posture, gait, occlusion, different weather (sunny,rain,snow), and lighting conditions. However the images in the video or significantly pedestrian is often in the complex environment, so that the human body detection and tracking of the human body become one of the most difficult challenges in the field of computer vision research.In this paper pedestrian detection and tracking which encountered pedestrians posture, gait, shelter, lighting conditions and other issues proposes appropriate solutions, different weather (sunny,rain,snow,etc.) is the focus of this study,the ultimate goal is to establish a relatively complete system of pedestrian detection and tracking algorithm to meet the needs of real applications. The main contents of the study include the following aspects:(1) A rapid, efficient and accurate pedestrian segmentation algorithm. Based Haar-features and AdaBoost learning algorithm to train the classifier, the classifier trained to meet the basic practical requirements.(2) The pedestrian detection processes with different weather (sunny,rain,snow,etc). Will leveling training a strong classifier based on multi-level class Haar features and AdaBoost learning algorithm trained weak classifiers to achieve different weather (sunny, rain, snow, etc.) in case detection, and extends the practical application of this algorithm range.(3) Put forward the pedestrian recognition algorithm which is based binary PSO (Particle Swarm Optimization) to improve the accuracy of pedestrian recognition.(4) Kalman tracking algorithm based on the combination of the Mean-Shift tracking the target. This algorithm of the proposed multi-pedestrian tracking has important significance.In the paper using this method to test and analyse two-dimensional space motion simulation and indoor environment.(5) Pedestrian detection system based on the establishment of anti-collision system, the algorithm is practical. The pedestrian detection system based on Haar-like features and AdaBoost learning algorithm to establish the vehicle and the establishment of anti-collision alarm system, the establishment of a more complete set of pedestrian-oriented detection and anti-collision alarm system to meet the needs of real-world applications.By a large number of experimental data and test results show that the proposed algorithm can be obtained at the correct rate,false detection rate and detection speed that is a complex, integrated, real-time and the robustness of the results that is very good pedestrian detection and tracking algorithm.