Research Of Obstacle Detection Algorithm Based On Multiple Cues Fusion

As the number of private cars increases, the frequency of traffic accidents also increases. The problem regarding safe driving for the car has generally become one of the attentional directions. Obstacle detection technology which is one of core technologies in the intelligent vehicle has already been the key research object of research institutions in every country. Obstacle detection, not only can improve the identification ability of drivers on the surrounding environment, but also can provide the assurance of other driver assistant detection technologies (Motion Estimation, Collision Warning, etc.) being effectively processed. Therefore, to study obstacle detection technology is very necessary.In the obstacle detection, as shape and type of obstacles are unknown in advance, obstacle detection cannot rely on the assumptions of shape, color and other special assumed conditions and prior knowledge of obstacles with specific types, then the detected obstacles are incomplete, locations of obstacles and roads are not exactly estimated, and then the false warnings are easily caused or the relative speed and collision time can not be computed accurately. Consequently, based on the analysis of existing detection algorithms, this paper proposes a novel fusion based obstacle detection algorithm. The fusion algorithm is divided into two steps:multiple cues fusion based on a single frame and history fusion based on multiple frames.In the multiple cues fusion, firstly, an improved motion compensation cue is utilized to detect obstacles which violate the road plane assumption. Secondly, a novel image segmentation cue is introduced to increase the obstacle detection rate and decrease false detection rate. The segmented cues include position cue, projection cue and area cue. Lastly, Bayes framework is employed to the fusion of these cues to output a probability map of obstacles in current frame.In the history fusion, this paper proposes a probability updating model to the fusion of detection information of current and history. On the basis of different detection results in the current frame, the algorithm selects different ways to update probabilities, which effectively integrates detection results of the current frame and history frames.The experiment under various kinds of scenes shows that the algorithm is not influenced by the types of obstacles, illumination conditions, road scenes and other factors. Meanwhile, the algorithm improves the detection rate and decreases the false rate on the assurance of overcoming the incomplete problem of detected obstacles.