Research On Algorithm For Detecting Front Vehicles Based On Mocular Vision And Multi-Features

In recent years, with the development of the intelligent driving technology domestically and abroad, the algorithm for detecting front vehicles based on vision, which is considered as a key technology of intelligent vehicle system, has become a central topic of current research. The purpose of this dissertation is to search accessible areas for intelligent vehicle.Using multi-features such as the shadow, texture, gray symmetry, taillights and license plate to detect front vehicles, which is based on monocular vision, is the main content of this dissertation. Though by analyzing and comparing various algorithms on vehicle detection introduced by a large number of domestic and foreign literatures, some approaches are proposed to improve the performance of vehicle detection algorithm. Finally, the vehicle detection system is designed. The main work of this dissertation includes points shown as follows.Firstly, in order to obtain the binary image, a novel threshold segmentation approach combining local gray value statistics with Twice-OTSU algorithm is presented. The gray values of pre-established areas are calculated. When the statistical information is available, the threshold is calculated by the gray mean and variance directly. Otherwise the Twice-OTSU algorithm is used to recalculate the threshold.Secondly, with the threshold above, the gray-scale image is converted to a binary image which is processed by the morphological methods such as erosion and dilation. According to the shape of shadow, a rough extraction method for target areas is proposed. Finally, the regions of interest are selected based on the positional relationship between the vehicle and its shadow.Thirdly, the texture, gray-scale symmetry, taillights and license plate are considered as three important features of front vehicle. Hence a method integrating the three features to confirm the ROI is proposed in this paper. At first, the entropy of the ROI is calculated. Some areas such as the surface of rode are eliminated. Then those areas are removed with the value of the gray-scale symmetry lower than the threshold pre-defined. Lastly, the reserved areas are converted from the RGB color space to the HSV color space. The pair of taillights and the licence plate between the two taillights are searched. According to the results, the ROI is determined.The algorithm for detecting front vehicle is implemented with Visual Studio2008.Net integrated development environment and OpenCV library. The experimental results demonstrate that the proposed algorithm is effective for vehicle detection and satisfies the real-time requirement.