Golf swing sequence with a seven iron: a address, b backswing, c top of the swing, d downswing, e impact, f follow-through, g finish 

Contexts in source publication

Context 1

... full swing consists of a series of key poses: address, backswing, top of the swing, downswing, impact, follow-through, and finish, as shown in Fig. 1. Among the successive swing procedures, the uncocking motion occurs in the downswing — the interval from the top of the swing to the impact. Therefore, the motion segmentation used to detect the downswing phase in the input golf swing is required prior to the uncocking motion analysis. In the motion segmentation module, the system recognizes three key pose moments in order to segment the sequential IMU sensor data of the golf swing stored in the dataset and then detects the ideal uncocking moment in the downswing phase. The three key poses are the beginning of the takeaway (the first movement after the address position), the top of the swing, and the impact. The ideal uncocking moment is defined as the moment at which the left arm is about 30° below the horizontal in the downswing, which is defined as the interval from the top of the swing to the impact. To detect the key poses and segment the input golf motion, we suppose that the user ’ s initial pose is the address, and set the x vector of the rotation matrix from the IMU sensor on the forearm at the address as an initial vector. Then, the system sequentially compares the x vector of the forearm in the array-type dataset with the initial vector in order to find four key events (beginning of the takeaway, top of the swing, impact, and ideal uncocking moment) and to segment the swing motion (backswing and downswing) based on the following procedures in Table 2. The system firstly sequentially calculates the included angles between the initial vector (the x vector of the rotation matrix from the IMU sensor on the forearm at the address) and every stored x vector of the forearm. The included angle can be computed from the inner product of the two vectors, as described in Eq. (5). The system removes noise from a series of included angles by using a 1D Gaussian smoothing filter. Then, it detects the beginning of the takeaway based on the above procedures. The 1D Gaussian smoothing filter for the discrete sequential data is derived via the following ...

Context 2

... chosen coordinate system. They measured rotational motion and estimated translation in the link model of the golf swing. However, to accurately measure the position and rotation of each body part, the length of each body part and the initial angle of each joint in the artificial link model must be manually defined by users and by using an additional sensor, such as an inclinometer. In addition, an estimate of the rotation using only gyro sensors might be not reliable, because small bias errors will quickly result in drift errors in the estimation [10]. Many researchers [8, 12, 18] have studied the kinematic analysis of the golf swing by using optical motion capture, which is one of the systems most often used to detect fast-moving human body parts. They analyzed the fundamental geometric and kinematic characteristics of the hip and trunk movements in the golf swing based on the measurement of the trajectories of target points (markers) attached to the human body by using multiple preinstalled cameras. These systems provide high accuracy, complete freedom of movement, and the possibility of interactions among various actors with a higher computational cost. To exploit this system for human body part tracking, however, the resources necessary to acquire the multi-view image data are a capture room, a body suit, and camera equipment. The capture room must be large enough to allow recording from a large number of viewpoints at a sufficient distance. The moving person should wear a special body suit with markers on certain body parts. Also, multiple cameras should be installed in the capture room, with scrupulous attention to ensuring synchronization, a high frame rate, and a suitable resolution. Therefore, expertise is necessarily required due to such issues, so it is difficult to use the optical motion capture for commercial and practical purposes regarding golf swing analysis, especially for amateur golfers who want to utilize this system personally. Particularly, the above techniques do not provide detailed feedback to help novices improve their swing skills; they focus on the quantitative analyses and measurement methodologies of the given golf swing. In this paper, we present an IMU sensor-based autonomous kinematic analysis platform for wrist angle measurement that enables the evaluation of a user ’ s uncocking motion in his or her golf swing. As mentioned above, an IMU sensor provides accurate orientation or rotation information so that it can be used for motion capture technology [13]. Because an IMU sensor can sense and transmit the orientation data with a high sampling rate, it is well-suited to the analysis of continuously and unobtrusively fast-moving human motions, especially to sports technologies for technique training [25]. In [6, 7], Ghasemzadeh et al. presented a golf swing training system that incorporates five custom-designed IMUs to obtain information on wrist rotation in relation to the direction of the club face and provide feedback on the quality of movements. To evaluate wrist rotation, they built a quantitative model by using PCA and LDA techniques for dimension reduction and the linear projection of data from sensors. However, the initial parameters for PCA and LDA, such as the number of principal components, must be defined carefully. If not, the evaluation method may be not reliable. Also, their system provides only numerical feedback. Compared to this system, the method proposed in this paper focuses on the wrist uncocking motion, which is to release the bended wrist at impact moment to generate additional power. This system can analyze and evaluate the uncocking motion automatically, without any initial parameters. Moreover, instructional and more specific feedback for the improvement of the user ’ s golf skill is provided. Table 1 shows the various kinds of representative techniques for motion analysis, as well as their limitations. Our system uses only two IMU sensors to measure the wrist angle quantitatively and has the ability to segment the inputted golf swing motions by recognizing the key poses and to detect the uncocking moment in the sequential swing motions. In order to help the user improve his or her skill, the system provides detailed verbal and graphical feedback based on the measurement result of the uncocking motion. Therefore, our system can be applied to a diagnostic system for swing correction by using multimodal feedback. In order to discern the important considerations in system development, the concepts of wrist cocking and uncocking in the golf swing must be examined. In this section, we briefly describe the considerations of our system with respect to the changes in wrist angle during the golf swing. Before beginning a detailed discussion of wrist cocking and uncocking, it is necessary to describe the golf swing. A full swing consists of a series of key poses: address, backswing, top of the swing, downswing, impact, follow-through, and finish, as shown in Fig. 1. In the process of these sequential motions, many factors, such as upper body twisting, wrist angle changes, weight shift, and so on, affect performance in terms of club speed and acceleration. Among these, the wrist angle, the relative angle between the forearm and golf club, is one of the most important factors in achieving accurate impact and distance. There are two representative motions in golf swing wrist angle, cocking and uncocking, as illustrated in Fig. 2. Cocking occurs naturally from the waist up during the backswing. At the top of the swing, cocking occurs in the direction of the thumb. If cocking occurs not toward the thumb, but toward the back or palm of the hand, the clubface will be misaligned, making it difficult to hit the ball straight. The golfer can generate additional power at the moment of impact and cause the follow-through to follow the ball by keeping the wrist cocked during the downswing until just before impact and then uncocking. Doing so will also make it easier to hit the ball straight. Thus, cocking and uncocking add power to the swing. Our system focuses on wrist uncocking in a full swing. In order to implement the detection of the uncocking moment and an evaluation algorithm, we must consider two technical problems: how to recognize key poses and how to quantitatively apply the aforementioned theory to uncocking detection. In the former case, we utilize the three- dimensional rotation of an IMU sensor attached to a golf club. For the latter, our system detects and evaluates the uncocking motion based on previous studies, which have defined the role of biomechanics in maximizing the distance and accuracy of golf shots through qualitative and quantitative evidence [9]. According to such research, a golfer can maximize the distance of his or her drives by uncocking the wrist when the left arm is about 30° below the horizontal on the downswing. Based on this research, the system can detect and evaluate key poses and uncocking by using three-dimensional IMU sensor data. The next section describes the proposed system in ...