This seed project aims to develop an automatic means based on unsupervised machine learning, neural network computing, and computer vision techniques to analyze video content filmed inside underground stormwater drains to help discover structural and functional related anomalies. Such anomalies may jeopardize the integrity of the overall protective infrastructure to maintain necessary slope safety from landslide, especially for the densely populated hillside areas in Hong Kong. The R&D methodology involves the use of deep learning methods to extract image features for vectorizing video imagery of underground drains. Because of unavailability of concrete knowledge of how damages or defects are presented inside different drains, unsupervised machine learning techniques are used to cluster all vectorized images into various groups. The resulting image clusters will then be visualized for identifying groups having damage-related issues and subsequently computationally prioritize the severity of the defective drains to determine the priority of remedial works. Hong Kong Housing Authority (HKHA) is one of the slope management users to survey structural integrity of underground stormwater drains beneath slopes surrounding around 130 hillside public housing estates. The project is planned to use HKHA’s drain videos as the basis for developing a prototype tool to help detect damages or anomalies. The resulting anomaly detection capability may help crossvalidate the consistency of manual survey results and the viewability of manually captured drain videos. The R&D work may lead to more opportunities to apply such anomaly detection video analytics to help strengthen quality-related issues from surveying video data of underlying facilities for preservation and safety purposes.