Study On Multi-Information Fusion Based Road Region Detection

To find passable region is an important issue for navigation technology. Nowadays, the achievement of many applications such as mobile robot, autonomous vehicle and some navigation devices for visually impaired person highly depends on solution of this problem. Researchers from a variety of domains devote large energy and time to the topic. This field is not only of high scientific value, but also will bring huge economic benefits in the future.Usually human beings or animals make use of information from several senses in their daily tasks. It is proven that making a decision using sensors together gets more accuracy and robustness than using them individually. Motivated from the theory, this thesis has done some researches on multi-information fusion based road scene recognition, the main contributions are as follows:1. Design and manufacture a hardware platform of intelligent guide system,which is based on stereo vision and inertial measurement unit. The hardware platform is light, concise,with friendly man-machine interface and it can be equipped easily.2. Based on the hardware platform, we propose a multi-information fusion based road planar extraction method, which mainly improves the calculation of homography matrix. Fusing the data from cameras and IMU, the proposed method select the optimal matrix from the homography matrixs obtained from two sensors by decision rules. So it still works when the road is lack of features or features are mismatched. In addition, the thesis presents a algorithm that can be used to estimate the motion of camera and obstacle. Through the experiment, we demonstrate that our method are accurate and effective.