[M-GI31] Earth and planetary informatics with huge data management

[MGI31-03] Smoke detection in monitoring video utilizing optical flow variance

★Invited Papers

*Kazutaka Kikuta¹, Yuki MURAKAMI¹, Ken T. Murata¹ (1.National Institute of Information and Communications Technology)

We propose a system that detects smoke from monitoring videos using visual IoT technology. This system detects smoke in the image of a PTZ (pan-tilt-zoom) camera installed outdoors, moves the viewpoint of the camera to that position, zooms in, and acquires an enlarged smoke image. Early smoke detection is crucial for disaster prevention. The detection process employs optical flow and color discrimination. Optical flow is a suitable technique for the high-resolution, high-time-resolution videos produced by visual IoT. Optical flow is a method for detecting the movement of objects from luminance information between image frames, but when applied to outdoor monitoring images, it detects moving objects such as cars in addition to smoke. In this research, we evaluate the spatial and temporal dispersion of optical flow vectors from multiple video frames to reduce non-smoke moving objects and noise, enabling accurate smoke detection. This is based on the observation that the area and direction of smoke movement remain constant in videos taken over several seconds. By operating the PTZ camera to the acquired smoke position, acquiring higher-resolution smoke images in real-time is realized.