Abstract:[Objective]To address the inefficiency, high cost, and poor real-time performance of medium-to-large inspection vehicles, this study proposes an edge intelligence-driven lightweight method for real-time road distress detection. [Method] The method utilizes a model named YOLO-Trip (enhanced with Triplet Attention and built upon YOLOv8s architecture) to efficiently extract chromatic and spatial features, integrated with TensorRT acceleration for edge-device deployment.To resolve odometry challenges, we design a self-calibrated high-frequency odometer combining IMU and GNSS data through Kalman filtering and linear interpolation. A low-power onboard edge computing platform is implemented to acquire and analyze pavement images autonomously. [Results] Experimental results show the odometry system achieves a maximum 0.4% sampling error deviation from wheel encoders at 0-40 km/h, significantly surpassing standalone GNSS solutions. The YOLO-Trip model exhibits 4.23% higher Recall and 2.19% greater mAP@50 than baseline models, while reducing parameters by 12.58% and computations by 8.45%. The system enables real-time detection of transverse cracks, longitudinal cracks, alligator cracks, and potholes with precise geotagging, validated on both rural cement roads and mountainous asphalt pavements, providing critical data support for road maintenance.