Advanced YOLO-DeepSort-Based System for Drainage Pipeline Defects Intelligent Detection

Abstract

The drainage pipeline network constitutes a critical component of municipal infrastructure. Effective inspection of drainage pipelines is crucial for maintaining urban functions. However, manual inspection methods are inefficient and prone to error. To address this, an intelligent system for detecting, tracking and managing defects in drainage pipelines using YOLO-DeepSort (YOLO-DS) is proposed for automated defect information collection. Initially, YOLOv7 is employed to train the defect detection model, achieving a mean Average Presion (mAP) of 91.1% and processing at 172 frames per second (FPS). Subsequently, the YOLO-DS detection-based tracking algorithm is utilized to track defects in the pipeline video. Moreover, within the tracking framework, an innovative fusion module combining detection and ReID features enhances cross-frame matching robustness. Under conditions of lens rotation, jitter, and blur, the system achieves confidence levels exceeding 87% for potholes, 78% for misalignments, and 80% for obstructions, respectively, demonstrating strong robustness. Lastly, the detection and tracking algorithms are integrated into the information management platform. The management platform facilitates intelligent identification and counting of pipeline defects, and includes a professional communication module for automatic generation drainage pipeline health assessment reports, thereby streamlining manual inspection processes and saving assessment time.

Keywords

• Pipeline
• YOLO-DS
• defect detection
• defect tracking
• system