Task-Driven Video Collection

摘要

Vision systems are increasingly being deployed to perform complexsurveillance tasks. While improved algorithms are being developed to performthese tasks, it is also important that data suitable for these algorithms be acquired- a non-trivial task in a dynamic and crowded scene viewed by multiple PTZcameras. In this paper, we describe a multi-camera system that collects imagesand videos of moving objects in such scenes, subject to task constraints. The systemconstructs "task visibility intervals" that contain information about what canbe sensed in future time intervals. Constructing these intervals requires predictionof future object motion and consideration of several factors such as objectocclusion and camera control parameters. Using a plane-sweep algorithm, theseatomic intervals can be combined to form multi-task intervals, during which asingle camera can collect videos suitable for multiple tasks simultaneously. Althoughcameras can then be scheduled based on the constructed intervals, findingan optimal schedule is a typical NP-hard problem. Due to this, and the lack ofexact future information in a dynamic environment, we propose several methodsfor fast camera scheduling that yield solutions within a small constant factor ofoptimal. Experimental results illustrate system capabilities for both real and morecomplicated simulated scenarios.