SBP-YOLO:A Lightweight Real-Time Model for Detecting Speed Bumps and Potholes toward Intelligent Vehicle Suspension Systems

TL;DR

SBP-YOLO targets fast, reliable detection of speed bumps and potholes for predictive vehicle suspension. It advances YOLOv11n with a four-scale P2–P5 detection framework, GhostConv and VoVGSCSPC backbones, LEDH for efficiency, and a hybrid loss training strategy that includes $L_{NWD}$ and knowledge distillation, all augmented by strong data augmentation. Quantization to FP16 with TensorRT yields real-time performance (e.g., around $139.5$ FPS on Jetson AGX Xavier) while achieving high accuracy ($mAP_{50}$ ~86–87% and $mAP_{50-95}$ ~50–55%), demonstrating suitability for embedded road-condition perception in intelligent suspension systems. The approach offers a practical path to predictive, vision-based suspension control, with robust performance across varying distances, lighting, and adverse weather, and shows clear advantages over baseline YOLO variants in both accuracy and efficiency.

Abstract

Speed bumps and potholes are the most common road anomalies, significantly affecting ride comfort and vehicle stability. Preview-based suspension control mitigates their impact by detecting such irregularities in advance and adjusting suspension parameters proactively. Accurate and real-time detection is essential, but embedded deployment is constrained by limited computational resources and the small size of targets in input images.To address these challenges, this paper proposes SBP-YOLO, an efficient detection framework for speed bumps and potholes in embedded systems. Built upon YOLOv11n, it integrates GhostConv and VoVGSCSPC modules in the backbone and neck to reduce computation while enhancing multi-scale semantic features. A P2-level branch improves small-object detection, and a lightweight and efficient detection head (LEDH) maintains accuracy with minimal overhead. A hybrid training strategy further enhances robustness under varying road and environmental conditions, combining NWD loss, BCKD knowledge distillation, and Albumentations-based augmentation. Experiments show that SBP-YOLO achieves 87.0% mAP, outperforming the YOLOv11n baseline by 5.8%. After TensorRT FP16 quantization, it runs at 139.5 FPS on Jetson AGX Xavier, yielding a 12.4% speedup over the P2-enhanced YOLOv11. These results demonstrate the framework's suitability for fast, low-latency road condition perception in embedded suspension control systems.

Figures (10)

• Figure 1: SBP-YOLO framework with Backbone, Neck, and Head. The Neck fuses multi-scale features (P2–P5) from the Backbone and passes them to the Head for classification and regression
• Figure 2: Illustration of the Ghost module design and its impact on feature activation. (a) Ghost module; (b) GhostConv architecture; (c) Comparison of backbone activation heat maps at the 5th and 7th convolutional layers.
• Figure 3: The flowchart of the VoVGSCSPC
• Figure 4: Structure of the proposed LEDH. Multi-scale features (P2--P5) are transformed by two stacked $3\times3$ GroupConv layers and then fed into parallel $1\times1$ convolution branches for classification and regression.
• Figure 5: Images of potholes and speed bumps in the dataset