基于固定数采样法的人体行为模式分类方法研究

Classification of Human Behavior Patterns Based on Fixed Number Sampling Method

在视频序列中信息量较大时,当前人体行为模式分类方法存在分类效率低下,分类误差较大的弊端。通过对目标轮廓信息的分析和处理,获取人体目标轮廓精确的位置信息并建立坐标系,在质心-边界距离法对人体轮廓进行描述的基础上,通过固定数采样法平均选取轮廓像素点,消除不必要的像素点,对轮廓像素点的选取进行优化,生成更加准确的质心-边界距离描述子。在人体行为模式分类中,首先使用前期数据进行学习,生成一系列的行为数据集,再通过本文的固定数采样法筛选得到的轮廓点,生成质心-边界距离描述子,与行为数据集中的数据进行相似性度量,得到行为识别结果。所设计方法大大降低了分类的时间,并且提高了识别的准确性;实验证明本文的方法能够对人体行为模式进行较好、高快地识别与分类。

When the amount of information in the video sequence is large,the current human behavior pattern classification efficiency is low and the classification error is large.Through the analysis and processing of the target contour information,the precise position information of the human body contour was obtained and the coordinate system was established.On the basis of the description of the human body contour by the centroid-boundary distance method,the contour pixels were averagely selected by the fixed number sampling method,the unnecessary pixels were eliminated,and the selection of the contour pixels was carried out.Optimize and generate more accurate centroid boundary distance descriptors.In the classification of human behavior patterns,a series of behavior data sets were generated by learning from the previous data.Then the outline points were filtered through the fixed number sampling method in this paper,and the centroid-boundary distance descriptor was generated.The similarity measure with the data in the behavior data sets was made to get the result of behavior recognition.The design method greatly reduced the time of classification,and improved the accuracy of recognition.It showed that our method can recognize and classify human behavior patterns better and faster.