| At present, the front vehicle detection methods include method based on vehicle’s characteristics, learning, optical flow field, etc. The method based on learning has the strong robustness, but it needs to select large quantities of vehicles and no-vehicle samples for training classifiers; the method based on optical flow field needs a large amount of calculation, and its real-time performance is poor; In addition, it is sensitive to noise and light; The method based on vehicle characteristics has become one of the most widely used vehicle detection for its good real-time performance, strong robustness and applicablity to the monocular vision system. This thesis uses the method to complete the front vehicle detection.In order to reduce the computational complexity, improve processing speed and separate the vehicle from the no-vehicle more effectively, many methods were put forward, such as choice of the preprocessing region, improvement of image threshold segmentation method and the comprehensive utilization of multiple characteristics of the vehicle, etc.Most people usually select the preprocessing region with the method of lane line detection. However, this method is only suitable for standardized road, such as highway, urban main road and so on. To make it be suitable for any road, this thesis has put forward and done a researche of a method for determining the preprocessing region based on virtual lane model. This method firstly builds a virtual lane with a certain length, width and height in road space, and then creates a lane model according to pinhole imaging principle, at last, marks the preprocessing area in the iamge accrodding to the translation relation between image coordinate system and camera coordinate system.This method has further narrowed the scope of preprocessing region, and has reduced the interference of useless information of vehicle feature extraction and shortens the processing time.About image threshold, this thesis has improved its deficiency on the iterative threshold calculation. In order to make the segmentation threshold more close to the grey value of the vehicle’s shadow area, this paper has introduced a weight coefficient to enhance the influence of low grey value of pixels points on the calculation of iterative threshold, and weakened the effect of the high grey value of pixels points on the calculation of iterative threshold, so as to achieve the purpose of reducing the final segmentation threshold.For the problem of high mistakenly identified and missed detection because of use of a certain characteristic of vehicle singly, this thesis uses the vehicle’s shadow feature, edge feature, texture feature and gray symmetry feature such many fusion method for detecting front vehicles.Finally, under the environment of Windows7and Visual Studio2008, a kind of front-vehicle detection software was designed using OpenCV2.4.3and MFC. This thesis has maked an offline experiment by means of a video with different road conditions using the front-vehicle detection software, the experiment results suggest that the proposed detection algorithm’s accuracy rate is as high as90%, the average per frame takes about53millisecond. By comparing other methods proposed in the literatures, the method has made a good performance at accuracy and per frame takes on average. |
