Multi-objective optimisation of surveillance camera placement for bridge–ship collision early-warning using an improved non-dominated sorting genetic algorithm

Bridges spanning navigable waterways face increasing risk of accidental ship impacts due to the growing volume of waterborne transport, usually resulting in fatalities and substantial economic losses. Computer vision-based ship detection using camera networks provides an effective and cost-efficient solution for collision avoidance warnings. Although advanced algorithms have improved the robustness of visual systems under complex conditions such as night-time and atmospheric interference, their performance is still largely constrained by suboptimal camera deployment strategies. Determining an optimal surveillance layout remains challenging given the large-scale monitoring area and on-site installation constraints of bridge waterways. This study proposes a multi-objective-based camera placement framework integrated with an efficient optimisation approach to address this issue. Specifically, an improved Non-dominated Sorting Genetic Algorithm III (NSGA-III) is developed to reduce run-time complexity by eliminating redundant computations and incorporating adaptive memory matrices. A multi-objective function is designed to maximise camera coverage, enhance ship detectability, and minimise overall deployment costs. The effectiveness of the framework is validated through simulation-based experiments conducted on the waterway beneath a real-world long-span bridge. Two scenarios with different camera densities are explored. Compared to the standard NSGA-III and NSGA, the improved NSGA-III achieves higher computational efficacy and lower memory usage, leading to more effective camera deployment schemes. The optimised visual security systems are presented in a three-dimensional proxy virtual environment, with demonstration videos available at: https://github.com/congliaoxueCV/Display. The system-generated images consistently enable effective ship detection by the standard object detection model under various conditions.