A recognition algorithm for behaviors with high similarities based on Kinect
Chenxiao Fan, Yan Zhou · 2016
Although there are a number of algorithms proposed to solve the problem of action recognition, there is a lack of efficient algorithms for recognition of behaviors with high similarity. A stable and high-efficiency algorithm for movement search and recognition with high similarity is proposed in this paper. The Microsoft Kinect is used to conduct the action search at first, then the related joint information is obtained using the depth and skeleton frame data and the stable feature value of the related data is extracted to describe the aim movement. At last, the features are studied based on the BP neural networks. The feature vector extraction method and the study of the neural networks is the key idea of the algorithm in this paper. The result turns out that this algorithm can recognize high similar actions with an efficient and stable performance. On the basis of improvement of the robustness, the average accuracy of this algorithm is higher than 95%.