Enhanced PTZ Camera Dispatch Scheme for 3-D Environments Based on Deep Reinforcement Learning

In recent years, pan-tilt-zoom (PTZ) cameras with horizontal, vertical rotation, and zoom capabilities have enhanced the flexibility of fixed cameras. However, the existing research on PTZ cameras has not adequately addressed key monitoring requirements, such as viewing angle and resolution pixels per foot (PPF), often resulting in skewed perspectives or insufficient imaging quality. This study focuses on 3-D environments and aims to optimize PTZ camera dispatch for maximum target coverage while minimizing the number of cameras used. We formulate the problem as a mixed-integer linear programming problem, which is known to be NP-hard, and propose an intelligent solution using fast-prioritized experience deep Q-network (FPE-DDQN). This method leverages self-awareness and iterative learning to improve decision-making by incorporating past dispatch experiences and an innovative reward function design. FPE-DDQN prioritizes cameras based on their coverage potential and dispatch cost, significantly reducing the number of cameras needed while accelerating the adjustment of rotation angles and focal lengths. Experimental simulations demonstrate that FPE-DDQN achieves near-optimal coverage and dispatch efficiency, with the performance gap to the optimal solution being less than 0.52%–4.7%, while requiring only 1/10 to 1/100 of the computational time compared to the optimal solution.