Estimate of Head Posture Based on Coordinate Transformation with MP-MTM-LSTM Network

Abstract

Fatigue monitoring can effectively reduce or even avoid traffic accidents. Head posture estimation is one of the focuses in the field of fatigue monitoring. In this paper, according to the coordinate rotation transformation and neural network theory, a method for predicting the change of head posture with sight-line coordinates is proposed. First, the coordinate rotation transformation theory is used to replace the head posture change amount with the coordinate change amount, and the first-order difference value of the sight-line point coordinate is obtained by the difference method. Then, under the unified Cartesian coordinate system, the MP-MTM-LSTM neural network is established with the input information of first-order difference value and the output information of coordinate change amount. The innovation of this method is that the Cartesian coordinate change is employed instead of the Euler angle transformation. In the model verification phase, the true value of the head pose is collected by the posture meter. The experimental results show that the absolute error between the predicted value and the true value estimated by the new method is less than 15%. In the field of fatigue monitoring, the proposed method can estimate the amount of head posture change effectively, which is suitable for the case where the head center point is not fixed.