Stop Block – Stormwater Culvert Blockage Management
The StopBlock component of the Smart Stormwater Management project aims to address this lack of real-time data on culvert blockage by using a combination of state-of-the-art artificial intelligence, computer vision, edge-computing and IoT techniques.
More specifically, a new AIoT device has been the developed to assess in real-time, continuously and automatically whether a culvert is visually (partially) blocked or clear. The smart device is built around a CCTV camera paired with an edge-computer based on an NVIDIA Jetson TX2 hosting an AI visually assessing the status of the culvert. The remote connectivity and data transmission are provided by the 4G network. The AI has also been optimized for latency (i.e. computation time) and power consumption. The complete system drawing only 9.1W on average, it is solar-powered, and a battery enable a 24/7 operation. Being completely autonomous, the system can easily be installed anywhere
The two steps of the culvert blockage detection algorithm. The first step detects culvert openings using YOLO v4. Those openings are regions of interest that are then processed by a ResNet-50 classifier determining the blockage status of each opening.