Abstract: In the process of using angle method to remove the reverse gravitational acceleration, data error can result in low accuracy of linear acceleration. To address this problem, a bi-directional removal of the reverse gravitational acceleration method based on data segmentation is proposed. Firstly, a data segmentation method based on stationary points is designed. The stationary points are used as the endpoint to segment the motion data to avoid inter-segmentation accumulation of error in the angular velocity integration process.Then, a bi-directional angle method improved by the angle method is developed. The reverse gravitational acceleration is removed in units of data segmentations. The accumulation error was generated in the angular velocity integration process. The influence of the accumulated error on the accuracy of linear acceleration is reduced by the bi-directional angle method. Using the Raspberry Pi with the six-axis sensor integrated by a three-axis accelerometer and a three-axis gyroscope, two data sets are made in different ranges. Accuracy comparison experiments of different algorithm are conducted on two sets. The experimental results show the effectiveness and superiority of the proposed method.