Mobile robot navigation based on multiple external cameras in crowded environment

Abstract

The navigation of mobile robots in densely crowded environments remains a significant challenge due to the inherent limitations of onboard sensors such as LiDAR and monocular cameras, including severe occlusion, restricted perceptual range, and high computational burden. These limitations often lead to incomplete environmental understanding, unsafe navigation behaviors, and inefficient trajectory planning. In this project, we propose a comprehensive and novel navigation framework that leverages a multi-camera system externally deployed in the environment to achieve holistic, occlusion-resistant perception and highly reliable robot motion control. Central to our approach is a Generalized Multi-View Detection (GMVD) algorithm, which incorporates a learnable adaptive view projection mechanism that dynamically adjusts camera parameters to correct for calibration errors and environmental changes. This is combined with an attention-based dynamic view fusion module that selectively integrates features from multiple viewpoints to generate a high-resolution bird's-eye-view pedestrian occupancy map. To achieve accurate and continuous robot localization, a novel marker-based method is introduced, where a specially designed visual marker mounted on the robot is detected and reconstructed in 3D space via multi-view geometry, significantly enhancing positioning accuracy and robustness. For path planning, we design a hierarchical strategy that integrates a global planner based on an improved A* algorithm—incorporating pedestrian flow statistics and density-aware cost functions—with a local planner that fuses Velocity Obstacles (VO) and the Dynamic Window Approach (DWA) for real-time, collision-free avoidance of dynamic pedestrians. Extensive experiments are conducted in both simulated environments and real-world settings, demonstrating that our framework consistently outperforms state-of-the-art methods across multiple metrics, including success rate, navigation time, path smoothness, and safety compliance. The results confirm the feasibility and effectiveness of the proposed system in enabling safe, efficient, and socially compliant navigation in complex and crowded scenarios.

Date of Award	15 Apr 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Adam Rushworth (Supervisor), Salman Ijaz (Supervisor) & Ahmed Nasr Abdelwahed (Supervisor)