为了提高行为识别中上下楼识别的准确率,提出一种利用重力方向位移变化对上下楼区分的方法。将手机坐标系的加速度数据转换为大地坐标系的加速度数据,再利用大地坐标垂直方向的加速度分量得出垂直方向的位移量,并将位移量作为垂直方向的特征值之一。水平方向的特征值、合成加速度的特征值和垂直方向的特征值作为分类器的输入。实验结果表明,在不计手机放置位置时上下楼的识别率提高了10%以上,因此该方法有效地提高了行为识别的准确率。
In order to improve the recognition accuracy of ascending stair and descending stair activities,a method that uses the displacement of gravitational orientation to recognize ascending stairs and descending stairs is proposed in this paper.Theacceleration data acquired from mobile phone is transformed into the acceleration data of azimuth coordination,and the displacement of gravitational orientation is achieved by means of acceleration component in gravitational orientation of azimuth coordination.The displacement is used as one of the gravitational orientation characteristic values.The characteristic values in horizontaldirection,resultant acceleration and gravitational orientation are taken as the input of classifier.The experimental results showthat the recognition accuracy of ascending stair and descending stair activities improves10%no matter where mobile phone isput.Therefore,this method can effectively improve the accuracy of activity recognition.