Study On The Technology Of Pedestrian Detection Based On On-board Vision Systems

With the improvement of industrial level and the development of economy in our country,cars have become the common means of transportation for people,and the traffic congestion and traffic safety problems are serious.Therefore,it is very important to develop a car safety assistant driving system,which is supported by the research on the pedestrian detection technology.Because of the different human posture,the background environment is complicated and changeable,and people interact with each other.At the same time,there may be occlusion problems,therefore,in the vehicle environment,the high accuracy real-time pedestrian detection technology has become the research target.The main research content of this paper is divided into pedestrian detection and tracking.Firstly,the background,significance and research status of pedestrian detection and tracking technology are analyzed.Second,the basic theories and methods of pedestrian detection and tracking are studied thoroughly;Methods and related algorithms were improved to implement pedestrian detection based on human body parts and pedestrian tracking based on particle filter algorithm and mean shift algorithm.This paper proposes a pedestrian detection method based on human body parts.First,the human body is divided into three regions,and image features are extracted using haar-like features.Then the full-body detector and the detectors of head and shoulder,trunk and leg are trained using the Adaboost algorithm.Then,using a full-body detector to generate pedestrian candidate target areas,using head-shoulder,trunk and leg detectors in the corresponding parts of the area for detection,to determine whether the target candidate is a pedestrian by the final analysis of the site detection results.The images and data obtained through simulation experiments show the detection rate of our method proposed is improved,compared with the detection method of a single full-body detector and Anuj Mohan et al's introduction of a human detection method for position combination(Voting Combination Classifiers).Due to the need for multiple-site detection,the detection time is increased,compared to a single full-body test,but it can be guaranteed in real time.This paper proposes a target tracking algorithm based on the combination of color features and texture features combined with particle filtering and mean shift.Firstly,the popular particle filter and mean shift algorithm in moving target tracking are studied.The algorithm uses a single color feature to establish the target model.When the target color isclose to the background,the target cannot be accurately tracked.As the tracking time becomes longer and there will be problems such as particle degradation.Pedestrian goals in the process of movement will cause the size of the contour in the image to change with the distance from the camera.The traditional algorithm generally only tracks the target with constant size.When the size of the pedestrian target changes,it leads to inaccurate positioning.Aiming at the above problems,in this paper,we uses color texture features to fuse the target model,embeds the mean shift algorithm into the particle filter algorithm,and uses the mean shift algorithm to cluster the particle sets of the tracking process,which increases the weight of the particles,to close to the true position of the target,so we can accurately track the target by using fewer particles,and then it can ensure the timeliness of the tracking.The urban roads with different traffic scenes were tested.The experimental images and data show that the improved algorithm presented in this paper has higher tracking accuracy than traditional particle filter algorithms and mean-shift algorithms,and can adapt to changes in target size due to the need to cluster particles.The image processing time per frame is increased compared to the mean shift algorithm,but it can satisfy real-time tracking.