Weighted Fast Dynamic Time Warping based multi-view human activity recognition using a RGB-D sensor

In this paper, a real time multi-view human activity recognition model using a RGB-D (Red Green Blue-Depth) sensor is proposed. The method receives as input RGB-D data streams in real time from a Kinect for Windows V2 sensor. Initially, a skeleton-tracking algorithm is applied which gives 3D joint information of 25 unique joints. The presented approach uses a weighted version of the Fast Dynamic Time Warping that weighs the importance of each skeleton joint towards the Dynamic Time Warping (DTW) similarity cost. To recognize multi-view human activities, the weighted Dynamic Time Warping warps a time sequence of joint positions to reference time sequences and produces a similarity value. Experimental results demonstrate that the proposed method is robust, flexible and efficient with respect to multiple views activity recognition, scale and phase variations activities at different realistic scenes.