SBP-YOLO:A Lightweight Real-Time Model for Detecting Speed Bumps and Potholes

2025-08-02

AI Summary

To address the challenge of real-time speed bump and pothole detection in predictive suspension control for new-energy vehicles, this paper proposes SBP-YOLO, a lightweight detection model. Built upon the YOLOv11 architecture, SBP-YOLO incorporates GhostConv to reduce computational redundancy, VoVGSCSPC for enhanced multi-scale feature fusion, and an Efficient Detection Head (LEDH) to improve small-object localization accuracy. It further integrates Normalized Wasserstein Distance (NWD) loss, knowledge distillation, and weather-robustness enhancement strategies. Evaluated on the Jetson AGX Xavier platform, SBP-YOLO achieves real-time inference at 139.5 FPS with FP16 quantization, attaining an mAP of 87.0%—a 5.8 percentage-point improvement over YOLOv11n. The model thus simultaneously optimizes detection accuracy, inference speed, and embedded deployment efficiency, demonstrating practical viability and state-of-the-art performance in real-vehicle scenarios.

Technology Category

• Computer Vision: Object Detection & Categorization
• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

With increasing demand for ride comfort in new energy vehicles, accurate real-time detection of speed bumps and potholes is critical for predictive suspension control. This paper proposes SBP-YOLO, a lightweight detection framework based on YOLOv11, optimized for embedded deployment. The model integrates GhostConv for efficient computation, VoVGSCSPC for multi-scale feature enhancement, and a Lightweight Efficiency Detection Head (LEDH) to reduce early-stage feature processing costs. A hybrid training strategy combining NWD loss, knowledge distillation, and Albumentations-based weather augmentation improves detection robustness, especially for small and distant targets. Experiments show SBP-YOLO achieves 87.0% mAP (outperforming YOLOv11n by 5.8%) and runs at 139.5 FPS on a Jetson AGX Xavier with TensorRT FP16 quantization. The results validate its effectiveness for real-time road condition perception in intelligent suspension systems.

Problem

• Detect speed bumps and potholes in real-time
• Optimize lightweight model for embedded deployment
• Improve detection robustness for small targets

Innovation

• Lightweight YOLOv11-based detection framework
• Integrates GhostConv and VoVGSCSPC for efficiency
• Hybrid training with NWD loss and distillation