抄録
This paper describes an efficient solution to a multi-camera view planning problem. We deal with the case where a relatively small number of active cameras track many persons for surveillance purposes. Planning camera view is therefore one of the important issues in realizing a competent surveillance system. In our previous paper, we formulated the planning problem as a combinatorial optimization problem. Since its search space is very large, we developed a multistart local search (MLS)-based planner with non-linear programming for initial solution generation. The method is efficient enough to be used on-line. We apply the method to the problem of selecting several focused area in the image sequence from a fixed camera, which simulates the multi-camera view planning problem. Considering the success ratio of CamShift-based tracking, our planning method is considered to be sufficiently effective.