Python
Aim:
Design and validate an end-to-end, real-time, robust pipeline defect detection and tracking system based on a lightweight high-performance object detector and detection-based tracking (DeepSort-style fusion), and integrate it into a defect information management platform.
Abstract:
Drainage pipelines require continuous inspection to prevent failures such as blockages, misalignments, and structural damage. Manual CCTV video inspection is time-consuming and prone to errors, making automated defect detection essential. This project proposes an advanced real-time detection and tracking system using the latest deep learning model to identify pipeline defects more accurately and efficiently. The model processes CCTV pipeline footage to detect and classify defects automatically and integrates a tracking mechanism to avoid duplicate counting across frames. The system aims to improve detection accuracy, increase processing speed, and support maintenance planning through automated defect reports, replacing the limitations of current manual and old approaches.
Proposed System:
The proposed system introduces an intelligent pipeline defect detection framework using YOLO11, an improved and lightweight deep learning architecture capable of achieving higher precision and faster inference speed than YOLOv7. YOLO11 detects defect types such as potholes, misalignments, and obstructions from CCTV video frames and integrates an enhanced tracking mechanism to maintain consistent defect IDs across frames, preventing duplicate counting. This approach supports real-time inspection, reduces manual workload, and increases accuracy in pipeline health assessment.
Advantage:
Higher detection accuracy due to improved feature extraction and multi-scale fusion.
Better robustness under camera shake, blur, low illumination, and rotation conditions commonly found in pipeline inspections.
Scalability and easy integration into existing pipeline monitoring and management platforms.
Lightweight model architecture, suitable for deployment on inspection robots and edge devices.
Reviews
There are no reviews yet.