基于特征的车辆目标复合探测方法研究

Vehicle detection algorithm based on characteristic knowledge

联合使用车底阴影、边缘信息、轮廓对称、灰度对称、水平线对称等车辆先验知识,提出一种基于特征的车辆检测方法。首先在图像中构建具有车底阴影特性的感兴趣区域;然后在感兴趣区域中分析车辆的边缘置信度,滤除其中平坦的背景区域;接下来利用轮廓对称或灰度对称定位潜在车辆;最后在潜在车辆区域中进一步分析对称性测度和水平线特性。只有具有充分的边缘、对称性和水平线特性的潜在车辆区域才被检测为车辆。实验结果表明:该方法在城市环境且不考虑遮挡的情况下,车辆有效识别率达到90%以上。

A vehicle detection algorithm based on knowledge in image sequences is proposed for driver assistance system. Images are acquired by a camera installed in a moving vehicle. Prior knowledge of the vehicles, such as shadows underneath the vehicles, rich edge information, contour symmetry, gray level symmetry, horizontal line symmetry and etc, is combined. Firstly, the regions of interest in the image plane that exhibit the characteristics of the shadow projected underneath a vehicle are selected. Next the edge confidence is analyzed in the regions of interest ; contour symmetry or gray level symmetry is used to locate the potential vehicles. Symmetry measure and horizontal line feature of the regions of the potential vehicle are further analyzed. Only those potential vehicle regions that contain enough edge confidence, symmetry measure, symmetric horizontal lines are identified as vehicles. The approach is suitable for detecting vehicles observed in frontal view or rear view. Experiment results illustrate that under environment of town, TPR is up to 90% without considering sheltering, which meets the application requirements.