Compatible Sewer pipe defect detection and estimation of its key characteristic with two different imaging system

There is recently a trend of applying computer vision for interpreting the inspection images or videos of pips automatically. In recent years, deep learning has obtained promising performance in various computer vision tasks such as image classification and object detection. Compared with conventional computer vision techniques, approaches based on deep learning are capable of extracting image features automatically and there is not much requirement of image pre-processing, which improves the accuracy and efficiency. We attack this problem by formulating it as anomaly detection method. There are several techniques for anomaly detection, our goal is to improve the state-of-the-art technique like CFLOW-AD and tailored them to satisfy our client’s need. Using an unsupervised technique for detecting anomaly enable the system to detect any kind of anomaly without expecting providing ground truth by our client. In addition, using unsupervised technique enable the system to easily apply in different imaging system since it is not relying on the labeled data.

Faculty Supervisor:

Chi-Guhn Lee;Chul Min Yeum

Student:

Partner:

JACOBB

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Toronto; University of Waterloo

Program: