Enhanced-YOLOv8n: depth for small-scale pedestrian detection research on learning algorithms

Detecting little persons in object detection is always a difficult task. To overcome this issue, an improved method dubbed Enhanced-yolov8n is developed based on YOLOv8n. To improve the representation of small-scale pedestrians, the system incorporates a feature extraction detection layer tailored exclusively for small targets. In addition, deep separate convolution and efficient multi-scale EMA attention algorithms are incorporated into the original C2f to reduce model parameters while improving feature extraction for small pedestrians. Furthermore, the technique introduces the MPDIOU loss function, which aims to reduce the distance between the upper left and lower right points of the predicted bounding box and the true bounding box. This approach improves the accuracy and efficiency of bounding box regression. Experiments show that the proposed approach achieves a mAP50 of 84.6% on the CrowdHuman dataset, with a detection rate of 20.2% on small-scale pedestrians—a 1.7% improvement over the baseline. Furthermore, on the WiderPerson dataset, the algorithm achieves a mAP50 of 77.4% and a mAP50-95 of 49.7%, showcasing significant improvements compared to classical algorithms. These results highlight the effectiveness of the Enhanced-Yolov8n algorithm in enhancing the detection capabilities of small-scale target pedestrians, making it a promising approach in object detection.