Improving 3D Active Visual Tracking

Tracking in 3D with an active vision system depends on the performance of both motor control and vision algorithms. Tracking is performed based on different visual behaviors, namely smooth pursuit and vergence control. A major issue in a system performing tracking is its robustness to partial occlusion of the target as well as its robustness to sudden changes of target trajectory. Another important issue is the reconstruction of the 3D trajectory of the target. These issues can only be dealt with if the performance of the algorithms is evaluated. The evaluation of such performances enable the identification of the limits and weaknesses in the system behavior. In this paper we describe the results of the analysis of a binocular tracking system. To perform the evaluation a control framework was used both for the vision algorithms and for the servo-mechanical system. Due to the geometry changes in an active vision system, the problem of defining and generating system reference inputs has specific features. In this paper we analyze this problem, proposing and justifying a methodology for the definition and generation of such reference inputs. As a result several algorithms were improved and the global performance of the system was also enhanced. This paper proposes a methodology for such an analysis (and resulting enhancements) based on techniques from control theory.