YOLO-CLA: An Efficient and Lightweight Dense Target Detection Model Based on YOLOv5

In modern society, with the acceleration of urbanization and increasing population density, pedestrian detection technology in crowded scenes has become particularly important in public safety and management. Accurately identifying and monitoring pedestrian movement not only enhances safety in crowded areas but also provides technical support for applications such as intelligent traffic management, disaster emergency response, and public facility planning. In recent years, advancements in deep learning technology have brought new opportunities for pedestrian detection. With efficient image processing and model optimization, existing algorithms have seen significant improvements in detection accuracy and speed in complex environments. However, traditional algorithms still face challenges in crowded pedestrian detection, such as accurately locating target individuals in scenarios with occlusion, where conventional deep learning algorithms struggle to pinpoint subjects accurately. This study aims to design an efficient pedestrian detection method based on an improved YOLO V5 to address the key challenges of pedestrian recognition in crowded scenes, enhancing the model's robustness and adaptability in complex environments. To tackle the difficulty of locating individuals in crowded scenes, we propose a module called CSP Bottleneck With Lightweight Attention. This module improves the shortcomings of the C3 module in YOLO V5 by introducing a lightweight attention mechanism that enhances the C3 module's ability to extract global features, helping the model understand the relationships between objects in the scene and enabling it to detect occluded individuals. Experimental results on the WiderPerson public database show that the algorithm achieved Recall, AP, and mMR scores of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1. 6 3 \%, ~} \mathbf{8 5. 3 3 \%}$</tex>, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 5. 6 1 \%}$</tex>, respectively. Compared with other object detection methods, this algorithm demonstrates superior detection performance.