Real-time multi object tracking from different datastreams at different frequencies

In many domains of medecine like rehabilitation or computer-aided diagnostic can an accurate motion tracking and analysis system be usefull. We’ll present here a method to track constrained objects in real-time with high accuracy given differents data streams at different frequencies. Suppose that accelerations of the objects to track are available in the object-related coordinate systems at high-frequency f1 and that their absolute orientations and locations are only available at low-frequency f2. Our algorithm is able to provide a good approximation of 2D absolute orientations and locations at frequency f1. To do that we combine double integration of the accelerations signal between locations and orientations observations with constraints on objects. These constraints can be of several types e.g. relative fixed orientations between objects, relative distances,... and permit us to approximate orientations and refine earlier computed locations. Simulations with f1 = 4 × f2 give good results with a large amount of motions if we have enough constraints between objects and validate thus the method. However a lot of work is still to be done to relax the constraints to non-fixed ones using probablility density functions for relative orientations and locations, to extend the algorithm to 3D and to adapt it to bodyparts motion tracking.