A novel method of shuttlecock trajectory tracking and prediction for a badminton robot

Abstract

This paper proposes an aerodynamic analysis of the shuttlecock and a novel method for predicting shuttlecock trajectory. First, we have established a shuttlecock track data set by an infrared-based binocular vision system. Then the unscented Kalman filter algorithm is designed to further filter the noise and visual recognition algorithm errors. Third, the radial basis function (RBF)-based track prediction model is designed. This method offers a concept to obtain the neural network model of different kinds of flying or moving objects. The experimental results show that the proposed method can predict the shuttlecock trajectory in real time at high accuracy and can be used for implementing the algorithm of return strategies in the near future.

Visual Perception and Multimodal Control: A Novel Approach to Designing an Intelligent Badminton Serving Device

“…In this context, the posture is identified as a "cross hand". Here, dx and dy are given by: dx = x 15 − x 16 (1) dy = y 15 −y 16 (2) dy y y (2) In the gesture detection segment, upon acquiring the coordinate information of the hand's key points, the system calculates the distance between the key points and the joint curvature of a single finger. Subsequently, customized semantic judgment is applied to recognize the gesture, achieving system-wide gesture recognition.…”

“…With the enhancement of individuals' material living standards, an increasing number of people are directing their attention toward the holistic development of physical and mental well-being [1]. Badminton, being an exceptionally dynamic and competitive sport, demands athletes to possess advanced serving skills, and the capability to serve with precision and variety is pivotal for securing an advantage on the court [2,3]. In conventional training methods, coaches are required to repeatedly demonstrate various serve techniques over an extended period.…”

Automatic Shuttlecock Motion Recognition Using Deep Learning

“…However, their classification relied solely on the gradient features of single-frame shot motion images using the Histogram of Oriented Gradients (HOG) technique, lacking motion information. Zhi et al [3] classified the main player's shot motions into forehand shots, backhand shots, kill shots, and other types but did not perform temporal positioning of the shot motions. In this study, we locate the shot motions of the main player (i.e., the player facing away from the camera, occupying the main region of the video) in shuttlecock singles match videos and classify them into four types: forehand shots, backhand shots, smash shots, and drop shots.…”