基于视觉及多特征的前方车辆检测算法

针对现有视频车辆检测算法受光照、噪声等环境因素影响大,漏检和误检率高,难以同时满足鲁棒性及实时性的问题,提出了一种完整的前方车辆检测算法.该算法在改进的Hough变换提取车道线的基础上,首先对图像进行自适应二值化处理,通过腐蚀、膨胀法滤除干扰点;使用简洁有效的方法进行阴影线的合并及ROI区域的提取;算法利用目标区域内的信息熵、车尾对称性特征对感兴趣区域(region of interest,ROI)进行筛选和判别,降低了算法的漏检和误检率;使用改进的Robinson方向检测算子提取车辆边界,取得了较好的效果.结果表明:在处理分辨率为640×480的视频时,检测正确率89%,运算速度平均为17.6帧/s.

Front Vehicle Identification Algorithm Based on Visual and Multi-feature

Existing algorithms of video vehicle detection can be affected by light, noise and other environmental factors with high missed and false detection rate, and it is also difficult to meet robust and real-time, a complete algorithm of front vehicle detection was presented. On the basis of the improved Hough transform extracting the lane, this algorithm first processed the image by adaptive binarization, filtering out interference point through method of corrosion and expansion. Shaded area was merged and the ROI was extracted in simple and effective way. This algorithm could utilize entropy and rear symmetry characteristics to screen and discern the ROI area, reducing the missing and false detection rate. Good results were achieved by using Robinson-direction detection operator to extract the boundary of vehicle. Results show that when the video has a resolution of 640 í480 pixels, the correct recognition rate has achieved 89.4%, and there will be 17.65 frame to be processed within per second.