基于改进密集轨迹算法的人体行为识别

Human Behavior Recognition Based on Improved Dense Rtajectory Algorithm

研究人体行为准确识别问题。在应用传统密集轨迹算法进行人体行为识别时,在长时间的跟踪轨迹过程中混杂着背景干扰轨迹,容易导致轨迹漂移问题。而在应用支持向量机分类器对不同的数据集进行分类时也存在着兼容性问题。为了避免以上弊端,提出了一种改进的密集轨迹算法。结合选择性搜索算法定位人体区域并计算运动显著性信息差异进行轨迹提纯,减少背景的干扰。利用改进的人工蜂群算法对非线性支持向量机的参数进行优化,并对提纯后的轨迹特征进行分类,得到人体行为识别的结果。实验结果表明,利用上述算法对不同的数据集进行人体行为识别都有优于其它经典算法的识别结果,说明该算法能够极大地提升人体行为识别的准确率。

Study the problem of accurate identification of human behavior. When the traditional dense trajectory algorithm is used for human behavior recognition, the background interference trajectory is mixed in the long-term trajectory tracking process, which easily leads to the problem of trajectory drift. There are also compatibility issues when using support vector machine classifiers to classify different data sets. In order to avoid the above drawbacks, an improved dense trajectory algorithm is proposed. Combined with the selective search algorithm to locate the human body area and calculate the significant difference of motion information, the trajectory was purify to reduce background interference. The improved artificial bee colony algorithm was used to optimize the parameters of the nonlinear support vector machine, and the purified trajectory features were classified to obtain the result of human behavior recognition. The experimental results show that using the above algorithm to recognize human behavior on different data sets has better recognition results than other classical algorithms, which shows that the algorithm can greatly improve the accuracy of human behavior recognition.