Sensor Location Optimization and Joint Angle Prediction in Hand Postures
Nayan Bhatt, Varadhan SKM · bioRxiv (Cold Spring Harbor Laboratory) · 2017
hand shaping during grasping is of interest in robotics, neuroscience, and biomechanics.In animation industry and biomechanics, tracking body and hand movements are essential.For tracking body movements and hand movements, a sizable number of markers are required.As number of marker increase complexity and cost also increase.Researchers have tried various approaches for selecting a set of minimum no. of markers for accurate reproduction of movements.Data-driven approaches have been proposed for classification of various grasps.Some studies have focussed on prediction of joint angles (Kang et al. 2012; Hoyet et al. 2012; Wheatland et al. 2013).Our objective is to predict joint angles from a reduced set of sensors.For preliminary sorting purpose, we used PCA based sensor ranking which is discussed further.Figure 1: Block diagram representation for sensor selection and joint angle prediction approach.The general block diagram for sensor selection and joint angle prediction is shown in figure 1 above.It shows 16 sensors need to be reduced to the predefined threshold value of 8. Three different algorithms were used for ranking the sensors.Our systematic approach combines these algorithms and gives possibly best reduced set of sensors.These reduced set of sensors were used for predicting all 21 joint angles.All the blocks are discussed in detail in future sections and subsections. PCA ranking based approachFor the purpose of analysis and ranking, all static postures were considered.Raw data from the sensors, the orientations (yaw, pitch, and roll) are used for computing importance of sensors and ranking all the sixteen sensors.First, we performed PCA on the dataset with 35000 samples