Performance Evaluation of the Azure Kinect Body Tracking Algorithm for Multi-Speaker Tracking

Microsoft has developed a body tracking algorithm that complements one of its products, the Azure Kinect. One of its features is the ability to track multiple humans in a video sequence using visual and depth data inputs. Currently, no study has been done on multi-speaker tracking using the data captured by Azure Kinect. We believe that there is a potential in the device for multi-speaker tracking that is yet to be discovered. Therefore, we capture the movement of multiple speakers using the Azure Kinect device and evaluate the tracking performance to verify whether it is suitable for multi-speaker tracking. The performance evaluation uses the multiple object tracking accuracy (MOTA) performance metric. The results reveal that the device and the tracking algorithm are suitable for multi-speaker tracking, and modifications are proposed to improve the tracking performance.