Visual tracking with sensing dynamics compensation using the Expectation-Maximization algorithm

Advances in vision-based technologies allow robots to perform sophisticated and intelligent tasks. Even with these advances, there still remain inherent problems with using vision-based technologies. Slow sampling rate and large latency is a problem associated with most vision hardware used in industry. We refer to these characteristics as the sensing dynamics associated with the vision sensor. This paper presents a compensation method that alleviates sensing dynamics issues in visual feedback tracking problems. We view the sensing dynamics compensation problem as two separate mathematical problems. Namely, we first deal with identifying the target model and then we deal with estimating the target position using the identified model and delayed measurements. The Expectation-Maximization algorithm and Kalman filtering are utilized to solve each problem respectively. The visual servo scheme associated with the proposed approach is also studied. Simulations and experiments are designed to test the performance capability of the proposed method.