Research On Road Detection Technology Based On Vehicular Image

With the rapid development of national transportation and road construction, the frequently occurrence of traffic accidents become a serious social problem. By capturing and analyzing the road information, automotive driver assistance system warn the driver of the vehicle’s departure from the driveway or possible collision. The automotive driver assistance technologies have great significance in avoiding traffic accidents and protecting people’s lives and properties. As one of the most important parts of such system, lane detection and road detection benefits further applications such as lane departure warning, automatic navigation and autonomous driving by providing important road information. Therefore, how to accurately and efficiently detect the lane and road is becoming a primary issue of the driver assistant system. This article focuses on lane detection and road detection algorithm based on image processing.In this thesis, we first introduce the existing research of lane detection and road detection. Based on the investigation of existing methods on lane detection and road detection, a robust lane detection algorithm and its hardware implementation are proposed in this thesis. We also propose a fast algorithm for road detection. The main contents of this thesis are as follows:1. In terms of the lane detection algorithm, we propose a lane information extraction method based on gradient magnitude and gradient direction, and the ROI adaptive segmentation based on gradient magnitude information. First, we calculate the edge map of the road image by using both gradient magnitude and direction. Then, the non-lane edges are suppressed by the gradient and the prior knowledge. The proposed algorithm minimizes the redundant information of the image and retains the lane features well. Therefore, it can improve the efficiency of the subsequent lane line fitting algorithm.2. In the lane extraction process, we improve the classic Hough transform by the way of grouping the gradient orientations, so that the target pixels are converted to parameter space more accurately. The improved Hough transform greatly improves the efficiency of the line detection. We carry out the programming from hardware architecture and software algorithms, so that the proposed lane detection algorithm can be computed on FPGA and DSP platforms in parallel way.3. For the research of the road detection algorithm, we use the illuminant invariance features and classification on similarity to solve the problem of the shadow interference to detection results. Moreover, we use the consistency between two images front and back. On the one hand, the optical flow method is used to obtain the location features of background area. On the other hand, we get the mean value of the invariance feature of the previous frame detection results. The proposed method is able to detect road region under different shadow situations efficiently and effectively.The lane detection and road detection algorithm evaluated in different image sequences, including shadows, cars, pedestrians and various road scenes, and the experimental results demonstrate its effectiveness and robustness.