Research On Road And Vehicle Detection Algorithm Based On Machine Vision

Visual-aided navigation is one of the hot issues on the navigation of mobile robot, in which the road detection and the vehicle detection are the important visual components. It combines image processing, pattern recognition, artificial intelligence and other related fields results, and has been widely applied in many areas, such as in video surveillance, robot navigation, intelligent car driving. Therefore, the subject has important theoretical significance and wide practical value.The paper starts from the structural design and software design of a monocular vision system, introduces the implementations of each function module, proposes the camera calibration technique, and discusses the parameters of the calibration in detail.Based on the images of roads and vehicles which are captured by the monocular camera, in order to improve speed and accuracy of the system detection and identification, an improved unstructured road edge detection algorithm and a dynamic vehicle detection algorithm are proposed respectively.For the detection of unstructured roads: first, use the median filter to filter the original road image, reduce the random noise; and then segment the filtered-images based on the Otsu multi-threshold method which is combined with the bimodal method, make the segmentation results and split time to be optimal; at last, use the Canny operator to detect the edges of road in the intersected images, after edge detection, then complete mathematical morphology amendment, and obtain the clear road edge image.For the moving vehicle detection algorithm: First, extract the edge of the road, obtain the drive area from the edge; reduce the search area of vehicle detection algorithm and confirm the region of interest in the light of experience of the drive area; then further narrow the region of interest based on the symmetry of vehicle, shadows, and edge; Finally, detect the moving vehicles on the filtered image by using the Adaboost classifier which is off-trained.Use the Visual C++ and the OpenCV function library to implement the algorithms, and the road and vehicle images are captured with the monocular camera. The simulation results demonstrate that the algorithm has good real-time, accuracy and robustness characters.