Multi-innovation anti-interference filtering method in face pose cooperative system

Abstract

The invention relates to a multi-innovation anti-interference filtering method in a face pose cooperative system, and belongs to the technical field of filtering. The multi-innovation anti-interference filtering method includes the steps: taking pose changes of a moving face and a standard face model as filtering observed quantities, introducing multi-innovation correction filtering estimation, and estimating state quantities of the face pose changes by means of multiple groups of observed quantities of a time sequence; judging filtering convergence and divergence in real time, estimating observation noise covariance and process noise covariance in time according to multi-innovation, and adjusting a Kalman gain matrix; and establishing a pose cooperation model, and calculating camera motion parameters according to the filtered face pose change to achieve cooperation of the camera and the face pose. According to the multi-innovation anti-interference filtering method, multiple groups ofobserved quantities of a time sequence are fully utilized, and system variance is estimated in real time, and Kalman filtering is adjusted, and noise is reduced to a greater extent, and response instantaneity is ensured while the stability of a face pose cooperative system is improved, and the multi-innovation anti-interference filtering method has good applicability to a system with high noise.

Description

technical field [0001] The invention relates to a multi-innovation anti-disturbance filtering method under a human face pose coordination system, and belongs to the field of filtering technology. Background technique [0002] Filtering is the key research content of almost all SLAM systems and robot autonomous navigation. Filtering mainly involves positioning, monitoring, estimation, search, exploration, navigation, manipulation, tracking, mapping, modeling and other fields. [0003] In order to solve the problem of signal processing and automatic control tracking in non-stationary environment, many filtering methods are adopted. For example, Kalman filtering is based on a linear time-varying system in a Gaussian environment. Given a set of observations, state estimation is done recursively. Or the particle filter approximates the probability density function with random samples propagated in the state space, and replaces the integral operation with the sample mean value t...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the problem of large noise in face pose calculation caused by camera shake in the existing face pose collaboration process, the present invention provides a multi-innovation anti-interference filtering method under the face pose collaboration system. Use the pose changes of moving faces and standard face models as filter observations, introduce multi-innovation correction filter estimation, and use multiple sets of observations in time series to estimate the state quantities of face pose changes; judge the convergence and divergence of filters in real time , estimate the covariance of observation noise and process noise in time according to multiple innovations, adjust the Kalman gain matrix; establish a pose collaboration model, calculate camera motion parameters according to the face pose change after filtering, and achieve the goal of camera and face pose synergy

Images