Jong-Sung Kim

Smart vision system for soccer training

This paper proposes a new smart vision system for soccer training. The proposed system adopts multi-exposure cameras to capture high-speed motion images of soccer ball instead of using high-tech or high-cost equipment. Then, new ball detection and motion analysis processes are adopted to detect the regions of soccer ball and analyze all motion parameters of soccer ball from those multi-exposed images of soccer ball in real time. The effectiveness of the proposed system is experimentally verified by comparing with a commercial ball motion sensor in terms of ball speed and spin rate. The usefulness and robustness of the proposed system is demonstrated by applying to an indoor soccer training platform in practice.

A new ICF-based ball tracking system with multi-exposure cameras for virtual sports

This paper proposes a new ball tracking system for virtual sports. In the proposed system, multi-exposure cameras are used to capture high-speed motion images of ball. Then, a new iterative circle-fitting algorithm based on the circular Hough transform, the OFF-cell filter, and the Hausdorff distance is adopted to track balls with unknown patterns from images. The effectiveness of the proposed ball tracking system is verified through experiments using real balls with various patterns. The usefulness of the proposed system is demonstrated by applying to a virtual sports platform in practice.

Articulated human motion tracking using a SOG-PF algorithm

This paper proposes a marker-free motion capture system to track articulated human motion using a new sum-of-Gaussian particle filter (SOG-PF) algorithm. The proposed motion capture system records multi-view color image sequences of human motion with synchronized multiple cameras. Then, the new SOG-PF algorithm that effectively combines 3D and 2D SOG models with a particle filter algorithm is adopted to track articulated human motion from multi-view color image sequences without visual markers. The performance of the proposed system is verified with real experiments.

Motion capture with high-speed RGB-D cameras

This paper proposes a new motion capture system to track human motion with high-speed RGB-D cameras. The proposed system uses synchronized and calibrated high-speed RGB-D cameras to simultaneously record color and depth image sequences of human motion. A new particle-based skeleton tracking process with color and depth features are adopted to track human motion without markers from high-speed RGB-D image sequences. The proposed system is very useful for digitizing performance.

A high-speed multi-exposure camera for virtual sports games

Virtual sports such as virtual golf are a sports game that is played with a real ball on a virtual environment. Most virtual sports games use a physics engine that simulates the trajectory of balls in flight with initial launch conditions, including ball speed, launch angle, spin angle and axis. Therefore, the measurement of initial launch conditions is essential for developing virtual sports games. A lot of systems based on speed sensors such as infrared LEDs have been studied for a long time. Nowadays, most commercial systems use high-cost equipment such as phased-array Doppler radars and high-speed video cameras. However, such high-cost equipment increases the system cost for developing virtual sports games. In the literature, several high-speed imaging techniques without high-cost equipment has been extensively studied for a decade. For example, multi-exposure cameras have been introduced to track the high-speed motion of a flying baseball or analyze the high-speed motion of human action. In this paper, a new high-speed multi-exposure camera is proposed to measure initial launch conditions of real golf balls for a virtual golf game without high-cost equipment.

A New Single Camera-based Ball Motion Analysis System for Virtual Sports

In this paper, a new single camera-based ball motion analysis system is proposed for virtual sports. In the proposed system, a ball motion imaging process using a single camera combined with a multi-exposure trigger is used to capture a ball motion image without high-cost equipment. Then, a 2D ball motion analysis process using pixel labeling and circle fitting algorithms is used to obtain the 2D ball positions and size from that ball motion image. Finally, a new 3D ball motion analysis process is used to simultaneously estimate the 3D ball positions and velocity by solving new ball motion equations based on some 2D-3D perspective constraints on the ball positions and size. The performance of the proposed system was experimentally verified against a multiple camera-based system and a radar-based system within a virtual soccer platform. Experimental results show that the proposed system is highly effective and efficient for virtual sports.

A robust vision system for outdoor swing analysis

In this paper, a robust vision system is proposed to detect balls with multi-exposure cameras to estimate ball speed and spin for outdoor swing analysis. A robust detection process based on adaptive thresholding and fitting techniques is used to detect balls from multi-exposed images regardless of shadows and reflections in outdoor environments. Experimental results show that the overall performance of the proposed system is comparable to that of a radar system for outdoor swing analysis.

Measuring initial conditions for baseball pitching games

Sports arcade games use a ball simulator to predict the trajectory of a flying ball with initial conditions, including ball speed and spin. Therefore, the accurate measurement of initial conditions of a flying ball is a critical issue for sports arcade games. In usual, high-tech equipment, such as Doppler radars and high-speed video cameras, is used for commercial sports arcade game systems. However, such high-tech equipment increases the system cost for developing sports arcade games. Multi-exposure cameras can be adopted for sports arcade games instead of using such high-tech, high-cost equipment. A golf game system with multi-exposure cameras has been introduced in the related work [Kim and Kim 2012]. In this paper, a new approach for measuring initial conditions of a pitched baseball with multi-exposure cameras is presented for baseball pitching games. The new approach can robustly detect baseball regions and surface patterns to measure initial conditions with high accuracy regardless of illumination conditions.