Video-based automatic tracking of three-dimensional human movement
The collection of kinematic data is routinely required for the biomechanical analysis of human movements. Available methods for obtaining kinematic data can be categorised into (a) direct methods, which are often limited by bulky instrumentation, and (b) imagebased methods. Current image-based methods generally necessitate the use of artificial body markers to aid the identification of body parts. A model-based method for the automatic tracking of human movement without the aid of body markers was developed. The approach constructed a three-dimensional (3D) computer graphics human body model that was customised to individual subjects via incorporation of subject-specific anthropometric data and appropriate colouring of model segments. Video image sequences of human movement were collected from multiple synchronised camera views. The environment from each camera view was simulated so that computer-generated model images containing the human body model could be matched to the associated video images. The human body model configuration was optimised through iterative adaptation of the model configuration in order to minimise the RGB colour difference between the model images and video images. A number of synthetic and video movement sequences were analysed using the tracking method. Synthetic image sequences of rigid and articulated motion were tracked with good accuracy. The tracking estimates obtained from video data of aerial movements were compared to estimates obtained via established procedures to provide an indication of the accuracy of the proposed approach. Movements that were successfully tracked returned estimates with errors comparable to manual digitising estimates. More complex twisting movements were tracked but with larger errors on all variables. The robustness of the tracking system was investigated through examination of tracking results following systematic perturbations made to selected tracking parameters. On both synthetic and real data the tracking estimates were found to be relatiyely robust to perturbations in camera and lighting parameters and reduced colour contrast. It was concluded that the tracking system presents a viable method for marker-free human movement tracking without representing a final solution to the problem.