Real-Time Defect Detection in Sewer Closed Circuit Television Inspection Videos

Closed circuit television (CCTV) is the most employed technology in inspection of sewer pipelines by municipalities in North America. Generally, visual inspection of sewer pipelines is done manually by a certified operator which is time-consuming, costly, and error-prone due to the operator's experience or fatigue. Automating the detection of anomalies can reduce time and cost of inspection while ensuring accuracy and quality of assessment. However, various types of defects in sewer pipelines and numerous patterns of each, make it difficult to detect the defects using computer vision techniques. This paper proposes an innovative and efficient anomaly detection algorithm to provide automated detection of sewer defects from data obtained from CCTV inspection videos. The algorithm employs Hidden Markov Model (HMM) for proportional data modeling. The algorithm performs real-time anomaly detection and localization and consists of modeling conditions considered as normal and detecting outliers to this model. The proposed model is tested on videos from sewer inspection report of the City of Laval, to evaluate its performance in anomaly detection in sewer CCTV videos.