Human activity recognition using motion trajectories

by 1972- Min, Junghye

iii A new method for extraction and temporal segmentation of multiple motion trajectories in human motion is presented. Motion trajectories are very compact and are representative features for activity recognition. The proposed method extracts motion trajectories generated by body parts without any initialization or any assumption of color distribution. Tracking human body parts (hands and feet) is difficult because the body parts which generate most of the motion trajectories are relatively small in relation to the human body. This problem is overcome by using a new motion segmentation method. At every frame, candidate motion locations are detected and set as Significant Motion Points (SMPs). The motion trajectories are obtained by combining SMPs with the color-optical flow based tracker results. But when a person’s location is not static (e.g. walking or jumping), hand or leg movements generate different motion trajectories which are recognized as different movements. A new method for the extraction of moving body part trajectories which are invariant to translational whole body movements is also presented which provides a more accurate description of activity. Using the HMM and the DTW, a performance comparison of activity recognition based on motion trajectories is performed. The experimental results show that the DTW is more appropriate for the activity recognition based on motion trajectories. The proposed temporal segmentation algorithm using the DTW is computationally efficient while successfully finding the start and the end point of each movement.