Abstract: Abstract: To address the issues of missed and false detections in dense pedestrian scenarios caused by complex backgrounds， high crowd density， low-light conditions， and partial occlusions， this paper proposes an optimized dense pedestrian detection algorithm based on YOLOv8n. The algorithm replaces the original convolutional blocks in the backbone network with efficient GSConv convolutions， reducing the model’s computational load while maintaining recognition accuracy. Additionally， GSConv convolutions enable the model to run efficiently on standard GPUs. The feature fusion network is replaced with the SlimNeck lightweight feature fusion module， which reduces the number of feature channels， thereby improving the model’s detection precision and speed. An EMA attention mechanism is embedded in the feature extraction network to enhance the model’s ability to capture both global and local information， thereby reducing false and missed detections in dense pedestrian scenarios. The algorithm also incorporates the Repulsion Loss function to better handle overlaps and occlusions among adjacent pedestrians in dense pedestrian detection， reducing interference between targets and optimizing bounding box regression. Training and validation on the CrowdHuman dataset demonstrate that the improved YOLOv8 model yields a 4.5 percentage points increase in mAP over the baseline. Furthermore, the model exhibits superior performance in dense crowds, occlusions, small-object detection, and low-light conditions, thereby offering an efficient and robust solution for dense pedestrian detection.

Key words: Key words: YOLOv8, complex scenarios, dense crowd recognition, repulsion loss function, object detection, Slim-Neck