Scholarlink

The number of surveillance cameras installed has been increasing significantly in recent years. Analyzing the surveillance images revealed a high incidence of people walking while on the phone (walking smartphone), which is a major accident risk. In this study, we developed a system to estimate danger levels for walking smartphone users. The system detected the walking smartphone users by prediction model based on CoAtNet and YOLOX. Furthermore, the system estimated the danger level for walking smartphone users from the detected results using fuzzy inference. The model inferred danger levels from the positions of walking smartphone users, the number of people in the surveillance camera images, and the distance between the user and the person. We confirmed that the danger level was low when there were no dangerous objects, such as persons and cars, around the user. By contrast, the danger level was high with dangerous objects, such as persons and cars, around the user. It could be realized to inform the walking smartphone users of danger before an accident occurs. These results will lead to a reduction in traffic accidents.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Danger Level Estimation for Smartphone Users Using Surveillance Camera Images with Fuzzy Inference

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Kentaro Mori,
  • Takashi Yoshimura,
  • Yuto Nakajima,
  • Kohei Naya,
  • Hideyuki Hayashi,
  • Takuya Baba

摘要

The number of surveillance cameras installed has been increasing significantly in recent years. Analyzing the surveillance images revealed a high incidence of people walking while on the phone (walking smartphone), which is a major accident risk. In this study, we developed a system to estimate danger levels for walking smartphone users. The system detected the walking smartphone users by prediction model based on CoAtNet and YOLOX. Furthermore, the system estimated the danger level for walking smartphone users from the detected results using fuzzy inference. The model inferred danger levels from the positions of walking smartphone users, the number of people in the surveillance camera images, and the distance between the user and the person. We confirmed that the danger level was low when there were no dangerous objects, such as persons and cars, around the user. By contrast, the danger level was high with dangerous objects, such as persons and cars, around the user. It could be realized to inform the walking smartphone users of danger before an accident occurs. These results will lead to a reduction in traffic accidents.