Print Email Facebook Twitter Designing an adaptable and low-cost system for gesture recognition using visible light Title Designing an adaptable and low-cost system for gesture recognition using visible light Author van de Water, Stijn (TU Delft Electrical Engineering, Mathematics and Computer Science; TU Delft Embedded and Networked Systems) Contributor Wang, Q. (mentor) Lofi, C. (graduation committee) Degree granting institution Delft University of Technology Programme Computer Science and Engineering Project CSE3000 Research Project Date 2022-06-22 Abstract This paper describes the design of an adaptable, low-cost, and energy efficient gesture detection system. The system leverages the ambient light available in the environment to perform Visible Light Sensing (VLS) using powerful Convolutional Neural Networks. The focus lies on designing a system that is capable of robust gesture detection in any environment while only utilizing a limited number (3) of photodiode sensors. The research conducted in this paper contributes to two crucial aspects of VLS gesture recognition systems. First, it shows how the photodiode sensors can be automatically fine-tuned using four resistors to achieve excellent sensing performance in a wide range of lighting environments from 50 Lux up to 150k Lux. Secondly it is shown that a photodiode arrangement consisting of an equilateral triangle with sides of 5cm facilitates the highest performance and robustness using Dynamic Time Warping for general and user-friendly gestures. Subject Visible Light SensingEmbedded systemsphotodiodeVLSEmbedded machine learningTinyMLPrinted Circuit BoardArduino Nano 33 BLELight IntensityOPT101Gesture recognitionAmbient light To reference this document use: http://resolver.tudelft.nl/uuid:d41f29df-8958-433a-8e83-defc292bb27f Part of collection Student theses Document type bachelor thesis Rights © 2022 Stijn van de Water Files PDF Research_Project_Stijn_va ... er_25_.pdf 5.04 MB Close viewer /islandora/object/uuid:d41f29df-8958-433a-8e83-defc292bb27f/datastream/OBJ/view