Detection Of Stationary Obstacles Based On Vision And Laser Information

With the increase of car ownership, traffic accidents have increased year by year. The situation has threated human's life and property. As car driving asssistance technologies can reduce traffic accidents and property damage, various countries have taken more attention on its study. The detection of vehicles, pedestrians and other obstacles in real time is one of most inpoatant contents of driving assistance due to that it can effectively prevent rear-end collisions, pedestrians injuries and other mailgnant traffic accidents.This paper based on the status of China's traffic and technical conditions and commited to detect and identify vehicles, pedestrians and other stationary obstacles in front of autonomous intelligent vehicles (this car). These test experiments were on the platform which is a smart car that develpoed by Dalian University of Technology. What's more, we can get the distance between the stationary obstacles and the vehicle according to the datas from laser so that the smart car can take measures to avoid obstacles on time.In this paper, we get access to information of environment in front of the vehicle in means of the monocular vision and laser sensor primarily. The research in this paper mainly carryies out detection and identification algorithm of stationary obstacles in preceding vehicle under normal light conditions.The spectific content is as following:(1) Because the traffic environment which pedestrians exist is complex, this paper proposes and implementes an effective method of pedestrian detection. The method first makes difference between the image of current frame and the image of background to segment the pedestrians from the complex background initially. As pedestrians accounts relatively small proportion of the entire image, we use the smallest difference error segmentation next. Then we use morphological algorithms in post-processing. At last, we combine the statistical pedestrian characteristics to recognise pedestrians. At the same time, first segmentation with minimum distance to segment pedestrian, then propose the criterion of pedestrian according to statistical analysis based on laser one-dimensional range image.(2) Under the basis of detection of vehicle's shadow, this paper proposes to use multi-threshold segmentation to eliminate the impact of penumbral of vehicles'shadow so that we can get relatively accurate region of interest of vehicles. Next we make use of texture, symmetry and other characteristics within the region of interest to exclude interference of the other obstacles. In the end, edge features are used to locate the vehicle preciselly. In the process of vehicle identification in terms of the laser range image, first data from laser is processed with median filter to eliminate isolatd points. Then region growing is used to segment region of interest of obstacle in front of vehicle. Last, the width of vehicle is used to determine whether vehicle is present.(3) In this paper, we take vehicle detection for example and research algorithm of multi-sensor information fusion. This paper first extractes rectangular degree of vehicle's shadow, texture, symmetry characteristic based on CCD image as well as line segment region information characteristics of obstacle in laser range image. Then we use the D-S evidence theory of data fusion method to carry on the vehicle to examine. Experiments show that the method has a high rate of vehicle identification, good examine system's robustness and reliability.