适用于遮挡车辆检测的子块带权模板匹配方法

Subblock-weighted template matching method applied to overlapped vehicle detection

针对多车辆跟踪过程中的遮挡问题,提出一种基于梯度方向直方图的子块带权模板匹配方法。该方法先对目标分块,并提取每块的梯度方向直方图,然后利用核函数为各块赋予不同权值,并采用子块带权特征匹配度度量方法计算目标模型与搜索窗的匹配度,进而获取最佳匹配。最后将该方法应用于多车辆跟踪过程中的遮挡车辆跟踪与检测。实验表明,该方法具有良好的精确度和鲁棒性。

To solve the occlusion problem in tracking of multiple vehicles,this paper proposed a subblock-weighted template matching method based on histograms of oriented gradients,which first divided the object into blocks and extracted histograms of oriented gradients for each block.By weighting different blocks into different grades using kernel function,this method calculated the similarity between the target model and the searching window based on the subblock-weighted feature matching degree measurement method,hereby,obtained the best matching region with the maximum calculation result.Finally,this paper used this method to tracking of the overlapped vehicle in multi-vehicle tracking process.The experiment result shows that this method embodies higher accuracy and robustness.