Scholarlink

In order to solve the problems of the installation work and the difficulty of understanding the thrust process in the electric referee machine (ERM) in the fencing foil event we have already constructed a new sword device that utilizes LEDs and advanced image processing. This device does not use an ERM, but instead lights up the LED at the tip of the sword at the moment of a thrust, and judges the valid thrust based on the hue information of the sword-tip peripheral image taken by the camera on the brim. In this device, a colored jacket is worn beneath the metal jacket on the torso, and the valid thrust is judged based on the hue information. However, the issue of hue discrimination remains, when wearing a normal white uniform. In this study, in order to distinguish the metal jacket from the white uniform we considered a method to use edge information caused by stripes on the metal jacket without using hue. In the process of the study, we encountered a problem that when a thrust is made quickly, the image becomes blurred due to the bending of the blade after the thrust, making it difficult to recognize the stripe pattern. Therefore, we considered a method to select the image with the least blur by evaluating the sharpness of multiple images taken at regular intervals after the thrust using edge information. We proposed a method to distinguish between metal jackets and uniforms based on the edge count of the selected image, and after testing by machine learning, we were able to achieve a classification accuracy of approximately 98% for judging valid thrusts. In addition, by applying a subtraction method between images with different thresholds, the classification accuracy was improved by approximately 2%, resulting in an accuracy of 100%. Moreover, an accuracy of 98% was possible with just three training data for preprocessing (a training time of approximately 10 seconds). This method was found to be effective in actual fencing match, as it reduced the preprocessing time. Moreover, by restricting the search range for the sword tip coordinates, a practical judging time of 0.42 seconds was achieved.

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

Development of a Light-Emitting Sword and a Device to Judge Valid Thrusts by Detecting Stripes on a Metal Jacket Without an Electric Referee Machine in Foil Fencing

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kaito Fujita,
  • Tatsuki Mori,
  • Seira Aguni,
  • Yumi Asahi,
  • Tetsuo Nishikawa

摘要

In order to solve the problems of the installation work and the difficulty of understanding the thrust process in the electric referee machine (ERM) in the fencing foil event we have already constructed a new sword device that utilizes LEDs and advanced image processing. This device does not use an ERM, but instead lights up the LED at the tip of the sword at the moment of a thrust, and judges the valid thrust based on the hue information of the sword-tip peripheral image taken by the camera on the brim. In this device, a colored jacket is worn beneath the metal jacket on the torso, and the valid thrust is judged based on the hue information. However, the issue of hue discrimination remains, when wearing a normal white uniform. In this study, in order to distinguish the metal jacket from the white uniform we considered a method to use edge information caused by stripes on the metal jacket without using hue. In the process of the study, we encountered a problem that when a thrust is made quickly, the image becomes blurred due to the bending of the blade after the thrust, making it difficult to recognize the stripe pattern. Therefore, we considered a method to select the image with the least blur by evaluating the sharpness of multiple images taken at regular intervals after the thrust using edge information. We proposed a method to distinguish between metal jackets and uniforms based on the edge count of the selected image, and after testing by machine learning, we were able to achieve a classification accuracy of approximately 98% for judging valid thrusts. In addition, by applying a subtraction method between images with different thresholds, the classification accuracy was improved by approximately 2%, resulting in an accuracy of 100%. Moreover, an accuracy of 98% was possible with just three training data for preprocessing (a training time of approximately 10 seconds). This method was found to be effective in actual fencing match, as it reduced the preprocessing time. Moreover, by restricting the search range for the sword tip coordinates, a practical judging time of 0.42 seconds was achieved.