Gesture Recognition in Leap Motion Using LDA and SVM
Hussein Walugembe, Chris Ian Phillips, Jesús Requena Carrión, Tijana Timotijevic · 2019
In this paper we propose a framework to support patients with hand injuries such as those recovering from strokes or similar illness to better perform daily tasks hence enhance their quality of life. This framework employs a low-cost “off-the-shelf” markerless sensor device called the Leap Motion controller (LM). During a rehabilitation procedure, we consider patients flexing and extending their fingers when their hands are placed above the LM. Various forms of measurement errors are expected including patients placing their hands out of reach of the LM “sweet-spot”. Experimental results show that this sweet-spot is always along the middle position of the device and within 25 cm elevation above its surface. During our evaluation, data collection was conducted using an “artist's hand” fixed above the LM when performing static gestures. This is because artist's hand gives more accurate measurements than a human hand as it can maintain a posture as long as is required. We have applied machine learning techniques such as Linear Discriminant Analysis (LDA) and Support Vector Machines (SVM). These techniques help in recognising and classifying performed gestures. In addition, these techniques can learn about measurement errors and compensate for them. Experimental results show a significant benefit when applying LDA and SVM yielding a performance accuracy above 88 %, which is far better than the baseline performance of 67.25%. Furthermore, as our approach enables more accurate measurements to be obtained, more meaningful feedback can be generated to help the patients that make use of the LM for hand rehabilitation exercises.