Sewer Defect Detection System

This repository contains the back-end code for the Sewer Defect Detection System as implemented at Royal HashkoningDHV. The system is developed to increase the efficiency and reliability of second-line cctv inspections of sewer infrastructure. It employs a deep learning object detection model (YOLOv5) to detect three types of defects (cracks, roots and displaced joints) and several filter methods to select the most relevant frame for each potential defect.

Installation

To install the system, clone this repository and install requirements.txt.

git clone https://github.com/jeannot-github/Sewer-Defect-Detection-System # clone repository
cd Sewer-Defect-Detection-System # change directory
pip install -r requirements.txt # install requirements

Preferably, you should also use CUDA and cuDNN for GPU acceleration.

Usage

The execute the system on new cctv footage, set the variables in config.yaml and execute main.py.

python main.py # execute 

Use the boolean variable LOCAL in config.yaml to load yolov5 locally or via the yolov5 Github (default is False).