**Abstract:** Pedestrian detection is crucial for applications such as intelligent transportation and video surveillance. While recent advancements in models like the YOLO series have improved performance, they still face challenges in handling diverse pedestrian scales and occlusions. To address these issues, the authors propose HF-YOLO, an advanced pedestrian detection model. HF-YOLO enhances feature fusion by leveraging both shallow localization information and deep semantic information, improving the detection of small-scale pedestrians and occluded instances. The model incorporates the HardSwish activation function to introduce more non-linear factors and strengthen feature representation. Additionally, a balance factor is introduced to the CIoU loss function to address regression imbalance, enhancing pedestrian localization accuracy. Experimental results demonstrate significant improvements, including a 3.52% increase in average precision, a 1.35% boost in accuracy, and a 4.83% enhancement in recall, while maintaining real-time performance with a detection time of 8.5ms. **Keywords:** Pedestrian detection · Object detection · Activation function · YOLO · Loss function**Abstract:** Pedestrian detection is crucial for applications such as intelligent transportation and video surveillance. While recent advancements in models like the YOLO series have improved performance, they still face challenges in handling diverse pedestrian scales and occlusions. To address these issues, the authors propose HF-YOLO, an advanced pedestrian detection model. HF-YOLO enhances feature fusion by leveraging both shallow localization information and deep semantic information, improving the detection of small-scale pedestrians and occluded instances. The model incorporates the HardSwish activation function to introduce more non-linear factors and strengthen feature representation. Additionally, a balance factor is introduced to the CIoU loss function to address regression imbalance, enhancing pedestrian localization accuracy. Experimental results demonstrate significant improvements, including a 3.52% increase in average precision, a 1.35% boost in accuracy, and a 4.83% enhancement in recall, while maintaining real-time performance with a detection time of 8.5ms. **Keywords:** Pedestrian detection · Object detection · Activation function · YOLO · Loss function