Binocular Vision-based Detection And Distance Measurement For Preceding Vehicle On Structured Roads

The accident rate increases significantly with the increasing of motor vehicles. Road safety has become a serious social issue. The DAS (Driver Assistance Systems) has attracted a great deal of attention lately to improve the driver’s vision efficiency and vehicle active safety performance. Preceding vehicle detection and distance measurement is an important component of DAS. It is used for detecting the front vehicle and measuring the distance between the driving car and the vehicle ahead and for giving alarms to alert the driver of the potential dangers. Vision-based vehicle detection and distance measurement has become a hot topic of DAS due to its’good features such as rich information, low cost, easy installation and reliable performance. Besides it doesn’t need to upgrade the existing infrastructure.In order to meet the requirement of real-time and robust performance, this thesis studies and improves detection and distance measurement algorithms for the preceding vehicle on the structured roads based on binocular vision. The main work is follows:(1) The initial region of interest (ROI) is defined based on the lanes by using the single edge single-pixel-width lane detection algorithm. The single edge single-pixel-width lane detection algorithm can detect the lanes in real-time and robustly. The ROI is then used to determine the existence of the vehicle. As a result, it avoids searching the whole image and saves the processing time.(2) The vehicle detection algorithm is studied and a preceding vehicle detection method is proposed by using the car shadow and Haar classifier. The vehicle candidate regions are determined in the initial ROI by using the shadow, which is detected by a shadow detection method. The vehicle candidate regions are then identified and verified by using the trained Haar classifier with high recognition rate. The experimental results show that the proposed vehicle detection algorithm meets the design requirements in time consuming and recognition rate of the vehicle.(3) The principle of binocular stereo vision ranging is studied and a preceding vehicle distance measurement method is given based on feature extraction and matching. In this thesis, the SIFT (Scale Invariant Feature Transform), SURF (Speeded up Robust Features) and ORB (Oriented Fast and Rotated Brief) algorithms are analyzed and compared. The ORB algorithm is chose finally due to its fast speed of feature extraction and the reasonable distribution of the feature points. This thesis eliminates mismatching points with polar constraint. Then according to the fact that the matching points’disparities are the same or closed to each other in the vehicle region, the mid-value of disparities is obtained and the non-vehicle disparities are abandoned according to the mid-value, which makes the ranging result more accurate. The experimental results show that the proposed distance measurement algorithm meets the design requirements in measurement accuracy and real-time performance.