Human Body Sequence Tracking using Unscented Kalman Filter with Implementation Details

In this project, we implement an Unscented Kalman Filter (UKF) for tracking human bodies in image sequences. The UKF algorithm is a nonlinear filtering approach based on Kalman filter principles, designed to estimate state variables including position, velocity, and acceleration. Our implementation follows the classical UKF algorithm structure, making it an ideal introductory program for learning UKF techniques. Key implementation features include sigma point generation using the unscented transformation and iterative state prediction-update cycles with weighted sigma points. We have enhanced the base algorithm with several improvements: integration of depth information processing through additional state variables and measurement models, and image noise correction using adaptive filtering techniques to improve tracking accuracy and stability. The code implements efficient covariance propagation and employs Cholesky decomposition for numerical stability during sigma point calculations. We believe this implementation has broad application prospects in areas such as human motion analysis, medical image processing, and computer vision systems, particularly for nonlinear dynamic systems requiring robust state estimation.