Coverage optimization of visual sensor networks for observing 3-D objects: survey and comparison

Abstract

Coverage is a basic and critical issue for design and deployment of visual sensor networks, however, the optimization problem is very challenging especially when considering coverage of three-dimensional (3-D) scenarios. This paper provides a brief survey of mainstream coverage optimization methods for visual sensor networks, including the greedy algorithm, genetic algorithm (GA), particle swarm optimization (PSO), binary integer programming (BIP) and differential evolution algorithm (DE). We provide an efficient open-source C++ implementation of these algorithms and compare their performance on a typical camera deployment problem for coverage of 3-D objects. In order to improve the computational efficiency, a parallel visual occlusion detection approach is proposed and implemented with graphic processing units (GPUs), which are then integrated into all of the aforementioned optimization approaches for a fair comparison. Evaluation results show that (1) the proposed parallel occlusion detection algorithm largely improves the computational efficiency; (2) among the five typical approaches, BIP has the best coverage performance yet with the highest time cost, and greedy algorithm is the fastest approach at the price of coverage performance; GA, PSO, and DE achieve a compromise between the performance and the time cost, while DE has better coverage performance and less time cost than PSO and GA. These results could serve as engineering guidelines and baselines for further improvement of coverage optimization algorithms.