Shoe Recognition Model with Floor Pressure Sensors (Poster)

2024年11月3日(日)〜7日(木)に開催されたSENSORCOMM2024の発表資料(ポスター)

Transcript

1. Shoe Recognition Model with Floor Pressure Sensors Sora Kamimura¹ Tetsuo Yutani² Atsuko Shibuya² Tsubasa Yumura¹ ¹) Hokkaido Information University ²) First Four Notes, LLC ・Currently, the mainstream of flow line analysis is a camera based method. However, blind spots caused by obstacles and privacy issues arise. ・These issues can be mitigated using floor pressure sensors for flow line analysis. ・There is a type of pressure sensor uses a pressure-sensitive conductive sheet Velostat. ・In flow line analysis using floor pressure sensors, it is necessary to identify the person from some factors. → We developed a method to identify shoe types using Velostat-based floor pressure sensors. Background and Purpose ・The structure of the sensor is shown in the figure based on our previous studies. ・The width of the copper foil tape was 5 mm and the spacing was 20 mm, and 12 tapes installed vertically and horizontally. ・This sensor is capable of measuring 144 points. Pressure sensor ・ In this study, we developed the neural network model to identify shoe type with pressure sensor with a Velostat. ・ Using a rotated dataset proved to be the most effective approach for real-world applications. ・there are large difference between the experimental environment and the assumed real environment, resulting in problems such as reaction rate and allowable pressure ・We will continue to develop both hardware and software to solve these problems. Conclusion ・The F-measure is generally higher for 10s data in both normal and rotational data. → This is because by waiting 10 seconds, vibrations and other noises when placed on table will subside. ・In 10 seconds of rotational data, the F-measure of each sensor is low, but F-measure of merged data is high. →The model could not learn each sensor's rotated data due to sensitivity bias, but in merged data, the amount of data has increased and the model can learn. Results and Discussion F-measure ・This model identifies three types of shoes: sneakers, room shoes, and sandals ・The model is a three fully connected layer neural network. ・The amount of data collected is shown in the figure. The data is categorized by the types of sensor, the number of seconds to measure, and whether the data was rotated. Recognition Model