摘要
In baseball, whether a batter has the ability to hit accurately is one of the key factors affecting his or her batting performance. Therefore, we want to develop a method for detecting the hitting position of the ball, using the vibration signal of the bat to estimate the collision position of the bat’s long-axis and circumferential direction. To achieve this goal, we collect the vibration signals of the bat through the accelerometer in the multi-axis inertial sensing unit.
Currently, in the prediction of long-axis collision locations, we collect data for different collision intensities. We extract the energy peak of each frequency as features. We use machine learning for training and estimation. Under single impact strength, the accuracy is close to 75%, and the tolerance error is within 1 cm. We use a low-intensity (1.1 J) impact to build a predictive model and estimate the location of a higher-intensity (1.73 J) impact. Near the sweet zone, the accuracy can exceed 60% with a tolerance error of 1 cm. What we propose is that it is not affected by the impact strength to a certain extent. In the prediction of circumferential impact positions, we utilize the characteristics of multi-axis accelerometers to identify methods for predicting different circumferential impact locations of a baseball bat. Under a single impact intensity, the accuracy is consistently close to 80%. However, when applying the model established with low intensity to predict higher intensities, the accuracy drops to only 50%. There still exists a challenge in accurately distinguishing impacts on both sides of the circumferential direction.