Dynamic Optimization Using Object Detection in Bandwidth Constrained Automatic Multi-camera Networks
Primary Investigator:
Yung-Hsiang Lu
Haobo Wang
Abstract
Large-scale multi-camera networks are getting an increasing demand for visual surveillance and multi-target tracking. However, there has been a challenge for the multi-camera surveillance system due to the limitation of bandwidth. Most recent researches on such system design consider either unlimited network resources, or under the worst case assumption which might lose information for uncovered areas. In this paper, we present a novel system to optimize the information obtained from multi-camera visual data given bandwidth limitations, which utilizes both the camera placement maximization and object detection algorithm. Our proposed method can define the priority of monitored cameras based on real-time video context and request different specs of data from individual cameras dynamically. We also suggest a cost-efficient, self-controllable multi-camera network emulation testbed, using minimega made by Sandia National Lab.