Adaptive visual object surveillance with continuously moving panning camera

In this paper, we study the important issues in the design of an efficient wireless visual surveillance system (WISE) in which a continuously moving panning camera is installed to capture real-time status of objects in a monitoring environment. To minimize the object evaluation workload, we propose a predictive scheme for evaluation on detected visual objects based on the movement of the objects. Due to movement of panning camera, visual object evaluation jobs may not be able to be performed following the prediction. In addition, overloading situations may occur due to the dynamic properties of objects in the environment. To resolve the problems, we propose two enhancements to the predictive evaluation scheme for evaluation in a panning camera, called Dynamic Multi-Resolution Analysis (DURA) and Load-Aware Job re-Arrangement (LAJA). DURA adaptively reduces the pixel-level processing cost for tracking while LAJA aims to balance the evaluation loading at the mobile computing device throughout the execution period by job re-arrangement after considering the object visibility in the panning camera. The proposed schemes have been implemented and their performance has been investigated through simulation. The results show that the evaluation workload is more evenly distributed and more interested events can be detected when both LAJA and DURA are applied at the same time. Copyright 2004 ACM.