Study On The Pedestrian Detection Based On Motin

The rapid development of automobile industry has brought great convenience to people's daily life, however, the resulting traffic accidents caused great damage to people's lives and property. So how to reduce the traffic accidents and the corresponding loss that they caused has become a hot topic. The vehicle driving safety assistant system as an important branch of new automotive technology, has attracted more and more people's attention because it can reduce the traffic accidents and protect the driver and the other road participants. Pedestrians, as the most vulnerable road users, are the member of the highest rate of casualties in the traffic accidents. In this case, the accurate preceding pedestrian detection has become the most urgent problem of vehicle driving safety assistant system.Although pedestrian detecting and tracking technology has been studied for years in the field of computer vision, However, there is not yet a common, robust, accurate, high-performance and real-time multi-target detecting and tracking algorithm, thus solving the problem of pedestrian detecting and tracking under certain limited conditions will still be the main topic in this field. At present, among the many sensors used for pedestrian detecting, vision sensors are widely used because it can capture a lot of useful information. Considering the accidents caused pedestrian deaths when crossing the carriageway accounted for more than 90 percent of total accidents which lead to pedestrian deaths. The main research work of this paper is about the pedestrian detection technology with a single stationary or moving camera, hope to resolve part of the problem existing in the vehicle driving safety assistant system and traffic monitoring system, what will lay the foundation for future research. The specific content is as following:1. In order to segment the moving objects in the image more effectively, the paper deeply research the method of image preprocessing and segmentation in the pedestrian detection system, including image enhancement, median filter, binary segmentation, morphological filter and so on. These processes ensure that we can extract a complete moving target in the follow-up work.2. For the pedestrian detection under static scene, the paper focuses on the improvement process of inter-frame difference method, finally using the method of five frame difference to extract a moving target completely, and experiments show that the algorithm is effective. 3. For the sake of segmentation of the moving targets under a moving background, the technology on the estimation and compensation about global motion is deeply researched. Comparing with the advantages and disadvantages about the existing movement model and global motion estimation method, we choose Six-parameter model to calculate the global motion, use a method based on SIFT features to describe the motion relationship between adjacent frames. After getting the feature points of the adjacent two frames, a method using Random Sample Consensus algorithm to get the global motion parameters is proposed, then we can complete the global motion compensation, So the problem is converted to detecting the moving targets under static scene. Experiments show that:the motion estimation algorithm based on SIFT features and RANSAC can perform well in this motion compensation.Based on Visual C++6.0, the paper develops the software and part of the algorithm, and the real image experiments have achieved satisfactory results, resolving part of the problem existing in the vehicle driving safety assistant system, providing technical support for future research.