Hand gesture recognition (HGR) is the process of identifying and interpreting hand gestures to control or interact with electronic devices. In this project, a Frequency Modulated Continuous Wave (FMCW) radar-based HGR is developed utilising Range-Doppler maps (RDMs). For this purpose, a Convolutional Neural Network (CNN) is implemented to classify different hand gestures. Each gesture that is fed to the network, contains a maximum of 12 frames, merged into a single image with a duration of 3 seconds. The dataset for training, validation, and offline test contains five different hand gestures along with Out-of-Distribution (OOD) samples, totalling 3235 data. The dataset was gathered in a confined environment with two participants, within a distance ranging from 0.2 m to 0.5 m. During training, the proposed system attained an accuracy of 95.91%, and 95.83% during training and validation, respectively. The system was also evaluated offline, achieving an accuracy of 96.99%. One objective of this project was to incorporate real-time functionality. In real-time testing, the system achieved 95% accuracy with a prediction time of 25 ms.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:mdh-67541 |
| Date | January 2024 |
| Creators | Haidari, Ihsan, Shen, Jiantao |
| Publisher | Mälardalens universitet, Akademin för innovation, design och teknik |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0016 seconds