Hand Gestures Recognition using Thermal Images

Abstract

Hand gesture recognition is important in a variety of applications, including medical systems and assistive technologies, human-computer interaction, human-robot interaction, industrial automation, virtual environment control, sign language translation, crisis and disaster management, en-tertainment and computer games, and robotics. RGB cameras are usually used for most of these applications. However, their performance is limited especially in low-light conditions. It is challenging to accurately classify the hand gestures in dark conditions. In this thesis, we propose the robust hand gestures recognition based on high resolution thermal imaging. These thermal images are captured using FLIR Lepton 3.5 thermal camera which is a high resolution thermal camera with a resolution of 160×120 pixels. Thereafter, we feed the captured thermal images to a deep CNN model to accurately classify the hand gestures. We evaluate the performance of the proposed model with the benchmark models in terms of accuracy as well as the inference time when deployed on edge computing devices such as Raspberry Pi 4 Model B and NVIDIA JETSON AGX XAVIER.