Presentation information

General Session

General Session » GS-10 AI application

[2F1-GS-10f] AI応用:安全

Wed. Jun 9, 2021 9:00 AM - 10:40 AM Room F (GS room 1)


9:40 AM - 10:00 AM

[2F1-GS-10f-03] Fallen Object Detection on Road by Using VAE Anomaly Detection with Average Image

〇Yoshiki Yamamoto1, Atsushi Hashimoto2, Yamato Okamoto1 (1. OMRON SOCIAL SOLUTIONS CO.LTD., 2. OMRON SINIC X Corporation)

Keywords:Social Infrastructure Application, Computer Vision, Anomaly Detection, Variational Auto-Encoder(VAE)

In order to keep roads safe and secure, it is useful to detect falling objects on roads automatically with surveillance cameras. Traditionally, the background subtraction method was used to detect falling objects. However, it sometimes detects environmental changes such as changing lighting conditions and shadows as falling objects. In this paper, we applied VAE (Variational Auto-Encoder) to falling object detection. In the experiment, compared with the background subtraction method of OpenCV, VAE showed better performance especially when the images including environmental changes. VAE increased the positive detection rate from 35% to 75% and decreased the negative detection rate from 15.0% to 2.4%.

Authentication for paper PDF access

A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.