Towards an Integrated System for Estimating Multi-joint Movement from Diverse Sensor Data

Motion capture has been an active research field for many years. Previous methods have mostly focused on converting segment orientations into joint angles. However such methods have a number of drawbacks: the estimates of joint angles are not optimal in a statistical sense; the common assumption that sensor placement is known and constant (relative to the limb) is in practice violated; existing methods are tailored to specific types of motion capture hardware, and cannot be easily adapted to new equipment or to situations where multiple sensor types are used simultaneously. This paper presents an approach that aims to resolve these problems, by combining modern statistical inference with domain-specific optimization. Our algorithm allows arbitrary collections of sensors to be used simultaneously (e.g. position and orientation markers, goniometers, gyroscopes), infers the sensor placements and segment sizes along with the time-varying joint angles (allowing the user to attach sensors quickly and inaccurately), and provides error bars for all quantities that it estimates