LP-YOLO: An improved lightweight pedestrian detection algorithm based on YOLOv11

• Published in: Digital signal processing Vol. 165; p. 105343
• Main Authors: Qu, Zenghui, Liu, Haiying, Kong, Weigang, Gu, Jason, Wang, Chaoqun, Deng, Lixia, Liu, Lida, Lin, Fei
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.10.2025
• Subjects: Deep learning; Lightweight model; Pedestrian detection; You only look once; Deep learning; Lightweight model; Pedestrian detection; You only look once
• ISSN: 1051-2004
• DOI: 10.1016/j.dsp.2025.105343

Pedestrian detection in complex urban environments poses significant challenges due to occlusions, scale variation, and background clutter. Given the growing demand for real-time inference on edge devices, the development of lightweight yet accurate detection models has become increasingly critical. This paper proposes Lightweight Pedestrian You Only Look Once (LP-YOLO), an improved neural network architecture based on YOLOv11n and optimized for urban pedestrian detection. The LP-YOLO model incorporates an enhanced Faster Implementation of CSP Bottleneck with Spatial Channel convolutions (C3k2SC) module to reduce redundancy in feature extraction across both spatial and channel dimensions, thereby enhancing feature representation. Additionally, the Spatial-channel decoupled Downsampling (SCDown) module facilitates lightweight downsampling, effectively reducing computational complexity and model size. Furthermore, an improved Feature Fusion Receptive-field Attention Convolution (FFRFA) module is integrated into the Neck to enhance feature fusion through the receptive-field attention mechanism, thereby improving small-object detection performance. Experiments on the CityPersons dataset show that LP-YOLO reduces parameters by 24.8% and computational cost by 7.94% compared to YOLOv11n, while improving mAP@0.5 by 1.6%. These findings demonstrate the effectiveness of the proposed approach in reducing computational demands while preserving or enhancing detection performance.