Research project on Template Matching (Training, Optimization and Evaluation). The TMM used is WarpingLCSS. The training is based on genetic algorithms.
WLCSSLearn is an algorithm for TMM training and parameters' optimization based on GA. It partially runs on CUDA, accelerating the training of WLCSS using GPGPU.
WLCSS is a TMM based on LCS that can be used on gestures warped in time. An implementation using CUDA framework is included in this project.
The evaluation of the training methods and of TMM are performed using several datasets, in isolated and continuous recognition.
[1] Mathias Ciliberto, Luis Ponce Cuspinera, and Daniel Roggen. "WLCSSLearn: learning algorithm for template matching-based gesture recognition systems." International Conference on Activity and Behavior Computing. Institute of Electrical and Electronics Engineers, 2019.
[2] Mathias Ciliberto, and Daniel Roggen. "WLCSSCuda: A CUDA Accelerated Template Matching Method for Gesture Recognition." Proceedings of the 2019 ACM International Joint Conference and 2019 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers. ACM, 2019.