A Human Gait Sequence Merging Method For Multi-kinect

Abstract

As a motion capture device, Kinect can predict temporal 3D positions of 25 body joints from depth image, so it is widely used in many field such as gait analysis, gait recognition and clinical medicine. However, the collection range of a single Kinect is very limited and it only collect few gait data, which will greatly reduce the accuracy and reliability of the gait analysis and recognition. In order to solve this problem, we propose a new human gait sequence merging method for multi-Kinect. It can not only extend the data acquisition range and increase the length of gait data sequence, but also avoid the bad effects of walking on a treadmill because it is a non-invasive, non-contact data collection method. Firstly, we first introduced the optimal collection range of kinect to improve the accuracy of gait sequence measurements. Secondly, we directly use the depth value of the joint point for coordinate transformation, which not only reduces the conversion error but also make the calculation easy. Finally, when gait sequences are merged, we utilize all the gait sequences to obtain stable, effective and long-distance gait sequences. We designed relevant experiments to compare the merged gait sequence with the gait sequence collected by single Kinect, and the results verified the validity of the method.