Gesture recognition with low quality signal over low energy Bluetooth 4.0

Accelerometer-based gesture recognition enables natural user interaction of handheld tools with PC and other devices. Dynamic time warping (DTW) can achieve high recognition accuracy and does not require extensive training. However, for applications with noisy data and limited data bandwidth, the performance of DTW algorithm decreases significantly. We propose a new gesture recognition method that is robust to low quality signals. We introduce a simple template selection step to obtain robust templates for DTW. Our experimental results show that our algorithm achieves an average 92% classification accuracy even under noisy and low bandwidth situations.