Department of Pattern Recognition

(eng): This Application Note aims to simulate a noise cancellation problem with MATLAB tools. This is purposed for pre-processing process for final gesture recognition application. It also shows advantages and disadvantages of an approach used for a noise cancellation. In applications for gesture recognition the signals can reflect and be detected not only from a desired source (a hand), but from the environment as well, which creates undesired noise and hardens the process of precise gesture identification. Therefore, it is essential to eliminate the signals, which come from other static sources than a hand. For these purposes echo cancellation methods can be used. Echo cancellation is widely and successfully applied in telephony in a way of preventing echo from being created or removing it after it is already present. We will assume that a hand will appear just for a short time period and the additional reflections will act as an additional short period “disturbance”. The echo cancellation in this specific case will be based on QRD algorithm with double precision arithmetic and exponential forgetting. The QRD algorithm is also called as an information filter without square root operations. It is based on QRD decomposition of the input/output information matrix. The recursively updated QRD factorization of the information matrix\nhelps to avoid the problem with loss of positive definiteness of the information matrix due to rounding errors and, thus, provides a numerically stable solution. The exponential forgetting is used instead of directional forgetting to keep the perspective of reduction of the computation time by applying the QRD version of the Lattice algorithm. QRD Lattice works only with the exponential forgetting with a constant exponential forgetting factor.