Design of a Lightweight Road Pedestrian DetectionSystem for Complex Environments

Yong Tang （ Hainan Vocational University of Science and Technology, Haikou, Hainan Province, 571126, China ）

Jun-Hao Zhang （ Hainan Vocational University of Science and Technology, Haikou, Hainan Province, 571126, China ）

Miao Sun （ Hainan Vocational University of Science and Technology, Haikou, Hainan Province, 571126, China ）

Abstract

To address the challenges posed by complex outdoor operating conditions—such as poor visibility, occlusions from rain gear, glare on water surfaces, and cluttered backgrounds—which cause traditional pedestrian detection models to suffer from parameter redundancy, a high deployment threshold, and imbalanced detection accuracy, this paper presents a lightweight road pedestrian detection system for complex environments, with a core focus on the deployment requirements for embedded terminals. By strategically selecting a baseline model, optimizing the feature extraction module, and streamlining the network architecture, and by incorporating strategies for rain-fog noise suppression and rain-glare reflection, the system strikes a balance between its computational load and recognition performance. The optimized system effectively adapts to a variety of complex road environments, providing technical support for pedestrian detection tasks on mobile devices under adverse weather conditions.

Keywords

Full Text

References