Road And Obstacle Detection Method Based On Information Fusion Research

The first work for the robot autonomous navigation is to detect and understand the surrounding environment. The robot can get and analyze the relevant environmental information from sensors on it, so the drivable region of the road can be detected for robot motion. Cameras and ladars are the most used sensors for robot autonomous navigation. The images obtained by the camera can describe the environment in a two-dimension plane, providing kinds of information of objects such as color, texture, geometry, etc; The ladar provides the distance information between the robot and objects. Different sensors provide information from different aspects, and effective fusion of these sensors information can make the description of the environment more accurate. The main work of this paper is to detect the road region and obstacles in front of the robot by using image and ladar data fusion.The mostly used methods for road detection are based on images. There are three ways for image-based road detection:road feature-based detection, road model-based detection and road region-based detection. In this paper we use the machine learning way for road region segmentation. We use Support Vector Machine as the classifier and extract the color, texture and edge features of the road region for classifying. Samples are obtained from road region and non road region to train the SVM(Support Vector Machine) classifier. After that the road image can be classified into two classes:road pixels and non road pixels. We train the SVM in-line, and update the samples and retrain the SVM classifier according to the classified results during the detection, so the algorithm can adapt the changing environment and get better results for road detection. As the SVM learning algorithm mentioned above requires human intervention, we improve the SVM algorithm of road detection. We use the ladar data to help robot get samples of the two classes by itself. FSVM(Fussy Support Vector Machine) are used instead of SVM to make sure that the samples are more credible and reduce the interference of the noise. It is proved that the image and ladar data fusion method can help improve the accuracy of the road detection algorithm.Obstacle detection is another important issue of the environment detection. In this paper we discuss the image-based and ladar-based obstacle detection algorithms in a complex background, and find that using the image only or ladar data only for the detection has highly missing rate of obstacles. So image detection results and ladar detection results are used together to verify the objects detected by each other. It is proved that this method can effectively reduce the missing rate.