Statistical Estimation and Adaptation for Visual Compensation in Object Tracking

The multi-modal tracking model in [ 1 ] enables the on-the-fly error compensation with low complexity by adopting acoustic sensors for the main tracking task and visual sensors for correcting possible tracking errors. The visual compensation process in the model is indispensable to the accurate tracking task in a dynamic object movement. This article proposes an algorithm to approximate the successful visual compensation rate appearing in the multi-modal tracking system. The acoustic sampling interval of the object signal and the random occurrence of transmission delays of multi-modal data are critical to the compensation process. Therefore, by using the two key factors as parameters, the algorithm called SEA can estimate the successful visual compensation of a tracking system. After we build up a tracking system, it is required to maintain the system at a certain level of tracking accuracy. This task can be done by controlling the aforementioned parameters since the visual compensation influences the tracking accuracy. Thus, we propose another algorithm called SEA 2 for the parameter adaptation. The algorithm controls only acoustic sampling interval due to the easiness of adjustment and having the main impact on the success of the visual compensation. From the algorithm validation, we show the SEA properly quantifies the visual compensation process successfully occurring in the tracking scenarios, and SEA 2 is feasible for parameter adaptation and achieving the target level of accuracy.