【摘要】 提出一种新的针对目标变形的模板匹配跟踪方法。对于跟踪序列图像中的目标,采用均值平移方法对原图像进行滤波平滑后提取边缘;定义一种点到邻域的广义距离测度,增强了匹配的容错性,与计算 Hausdorff 距离相比计算量大为减少;采用边缘相似点的距离均值和方差作为匹配的相似测度和置信准则,进一步降低了匹配的误差;提出在 8 邻域内基于马尔可夫模型的启发式规则修正模板策略,阻止跟踪点漂移。实验结果表明,该方法能以高达 98%的正确匹配率对复杂场景序列图像中的运动目标进行稳定跟踪。更多还原

【Abstract】 A novel approach for object tracking based on deformable template model in image sequence is presented in the paper. To track the objects in image sequences, the mean shift method, which filters the original image, is used to obtain good image edge. The generalized distance is defined between a point of template and its correspondence neighbors. Compared with Hausdorff distance calculation, this not only enhances matching fault-tolerant property, but its calculation quantity has also been greatly reduced. The matching error is further reduced through using distance mean value and variance of the edge similarity as similar metric of matching and rule of confidence level. The strategy of revising template model, which prevents template model from drifting towards background, is defined by the characteristics of pixel connection heuristic in terms of MRF model in 8-neighbors. The experimental results show that this algorithm can effectively and stably track target with 98% recognition rate in image sequences in complex scene.更多还原