YOLOv8-CB is a lightweight pedestrian detection algorithm designed for in-vehicle cameras, improving upon the YOLOv8n model. It introduces a lightweight cascade fusion network (CFNet) and a CBAM attention module to enhance multi-scale feature extraction and localization. The algorithm also integrates a bidirectional weighted feature fusion structure (BIFPN) to improve detection performance in complex environments. Experimental results show that YOLOv8-CB achieves a 2.4% increase in detection accuracy, a 6.45% reduction in model parameters, and a 6.74% decrease in computational load compared to YOLOv8n. The single-image inference time is 10.8 ms. The algorithm is effective for dense pedestrian detection in urban intersections and other complex scenes, offering a lightweight and efficient solution for device-side pedestrian detection with limited computational resources. YOLOv8-CB outperforms other algorithms in terms of detection accuracy, computational efficiency, and model size, making it suitable for real-time applications in challenging environments.YOLOv8-CB is a lightweight pedestrian detection algorithm designed for in-vehicle cameras, improving upon the YOLOv8n model. It introduces a lightweight cascade fusion network (CFNet) and a CBAM attention module to enhance multi-scale feature extraction and localization. The algorithm also integrates a bidirectional weighted feature fusion structure (BIFPN) to improve detection performance in complex environments. Experimental results show that YOLOv8-CB achieves a 2.4% increase in detection accuracy, a 6.45% reduction in model parameters, and a 6.74% decrease in computational load compared to YOLOv8n. The single-image inference time is 10.8 ms. The algorithm is effective for dense pedestrian detection in urban intersections and other complex scenes, offering a lightweight and efficient solution for device-side pedestrian detection with limited computational resources. YOLOv8-CB outperforms other algorithms in terms of detection accuracy, computational efficiency, and model size, making it suitable for real-time applications in challenging environments.