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

[MGI31-04] Moving object tracking by DeepSort in visual IoT system

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

Video IoT system, a type of IoT sensor, offers high-quality and real-time outdoor video capture. By utilizing a PTZ camera, it is possible to capture various areas with a single camera. However, since the position is fixed, the target detected in the image may be hidden by the shield. In this research, we perform stable detection of moving objects by taking vehicle monitoring of a coastal national highway as an example. A camera installed at an angle that captures the coastal national highway from the side is used, but cars in the back lane may be obscured by those in the front lane. Trees on the side of the road may also block vehicles. Continuous vehicle tracking was performed by using DeepSort, a type of AI image processing, and simultaneously determining the count and speed of vehicles. During tracking, we introduced a method to correct image blur caused by strong winds, which is unavoidable with outdoor cameras. Also, we created a learning data set for day and night, enabling traffic detection both during the day and night. The effectiveness was demonstrated by monitoring National Route 398 in Onagawa Town, Miyagi Prefecture.