| In recent years, with an increasing number of car ownership, traffic accidents occur frequently, which bring people a lot of personal injury and property damage. The increasing problems of traffic safety have caused the attention of people. Intelligent vehicle system has become hot spot of researching and the intelligent vehicle collision avoidance warning technology is becoming more and more i mportant. The detection of front vehicle is an important part of intelligent vehicle collision avoidance warning research. During the day ti me, the characteristics of the vehicle is more clear, method used for identification of front vehicle can realize and is accurate, but in poor light condition, with low visibility at night, the driver’s short line-of-sight more likely causes traffic accidents when driving, so the study of front vehicle detection at night has great practical significance.At present, the research of front vehicle detection at night mainly use monocular vision sensors, accessing to the information of visual perception, based on extracti ng taillight characteristics of front vehicles in image, but it can not get the depth information and need large amount of calculation, thus this paper put forward a detection method of front vehicle at night based on millimeter wave radar and machine vision, which uses the fusion information of multi-sensor to detect front vehicle. By millimeter wave radar to obtain the world coordinate of the vehicles and establishing the conversion relationship between the world coordinate system and image pixels, then converting the world coordinates of the radar target to image coordinates, and form ing the Region of Interest(ROI) on the image, and then using i mage processing methods to reduce the disturbance point of external environment, with D-S evidence theory fusing information of features and then t detect the vehicle. The main research content as follows:(1) The acquisition of effective target based on milli meter wave radar. Firstly to remove empty target signals, invalid target signals, static target signals. These three kinds of signals are interference. keep the targets within distance threshold and relative velocity threshold, and then according to the longitudinal width threshold to decide whether the target is in the lane or not. Finally according to nearest distance and the same lanes to select target effectively.(2) The fusion of radar data and machine vision information. First using radar to obtain candidate information such as the distance, angle, speed, and the world coordinate of the candidate target, and then reversing use camera calibration principle to get the transformation relations of world coordinates and image pixel coordinates. The initial candidate target is pointed to the image, namely the Region of Interest(ROI). Finally through the vehicle’s tail light detection method based on D-S evidence theory deciding whether there are vehicle’s characteristics in the ROI or not.(3) The detection of vehicle’s tail light based on D-S evidence theory. Extracting tail light feature information based on the collected i mages, with using D-S evidence theory to get the overall trust value fusing I feature information of vehicle. First of all, with using advanced OTSU threshold segmentation method segmenting image was carried out on the collected images, then processing the image based on image morphology and prior knowledge. After processing the i mage, tagging the connected area on the image. Finally, extracting the taillights characteristic information and using D-S evidence theory fusion information for detecting vehicles.Based on the software of VC++ 6.0 and the multi-sensor system, developing an algorithms of front of the vehicle detection at night. The experimental results show that the research using the millimeter wave radar and machine vision fusion method can effectively realize front vehicle detection at night and accurately positioning. It has good real-ti me performance and accuracy. |
PDF Full Text Request