An Improved Model for People Detection Based on YOLOv8

People detection is vital for security in large-scale disaster outdoor rescue and large public spaces. Despite its success in various applications, people detection technology continues to face challenges such as low accuracy, difficulty in feature extraction, and slow model performance. This paper proposes an enhanced people detection algorithm based on You Only Look Once version 8 (YOLOv8). The algorithm introduces the Cross Stage Partial Networks with Focused Expansion and Feature Fusion Compressed Information Bottleneck (C2fCIB) module to optimize the backbone network, enhance feature extraction capabilities, and improve the detection accuracy of small and occluded targets, thereby increasing the model's overall detection accuracy. Additionally, a combined loss function incorporating Skew Intersection over Union (SIOU) and Complete Intersection over Union (CIOU) is employed to optimize the detection rate for small and occluded targets, further enhancing accuracy and expediting convergence. Experimental results demonstrate that the average detection accuracy of the improved model increased by 15.4%, and the training speed improved by 15.5%. Consequently, the enhanced model is more effective for people detection, contributing to safety in public spaces and other application scenarios. Furthermore, the model's enhancement methods offer valuable insights for researchers and experts in the field of people detection.