救生衣-人系统漂浮运动测量中的标志点识别

Mark point recognition in motion tracking of a floating life jacket-man system

以曲率尺度空间技术为基础,提出了一种快速、高精度的圆形目标鲁棒识别算法。从边缘图像中提取目标轮廓的关键点作为轮廓的分割点,从中筛选出在一定准则下可接受的圆弧。经过两次聚类和两次拟合得到圆形目标的准确参数。对比实验表明,只要目标边缘的局部连续弧段足够长,该算法就可获得比Hough变换更高的精度,时间复杂度大大降低且不受目标尺寸的制约。该算法已用于人体水中漂浮运动跟踪,在复杂背景、光照不稳定和标志点被部分遮挡的情况下求取了人体运动的时间历程,获得了满意的效果。

A fast, high precision, robust algorithm was developed for round object recognition based on the curvature scale space technique. The algorithm first extracts critical points based on the curvature from the edge map to divide the object contour into several segments. An acceptable cireular arc is defined using statistical information of each edge segment. The parameters describing the round object are deduced using clustering and LS circle fitting. The algorithm has been evaluated with a number of images against the Hough transform （HT）. The results show that the method offers better precision than the HT when the lengths of the locally consecutive arcs are long enough and offers a far more effective solution with images containing widely different size objects. The algorithm has been used to track the motion of a floating man in a natatorium with complex backgrounds, variable lighting conditions and occlusion of marking points with very good results.