Real-time Baseball Detection and Three-Dimension Trajectory Reconstruction System With Unsynchronized Cameras

• Main Authors: Shang-Chih Lin, 林尚志
• Other Authors: Mon-Chau Shie
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/32993769059753778154

碩士 === 國立臺灣科技大學 === 電子工程系 === 98 === Pitcher plays the major role of baseball games. The skill of strike zone control is important for a good pitcher. Apply analysis system of video to pitcher’s training can reduce training time and lower injuries of sports. Apply this to a real game helps judgment of umpire. That's why this kind of systems is attracting people. In MLB, they use K-Zone to doing these jobs and other games use similar systems. We try to implement such a system for the purposes we mentioned before with inexpensive cameras. In order to implement it, we use inexpensive cameras with image capture cards and two laptops which connected by each other via Ethernet interface. Home-made External signal lighting device is used for synchronization. Here, the priority of the program will be raised to the real-time level to reduce the CPU time of context switch. Calibration is done by using checkerboard pattern in order to get the camera’s parameters. Because the moving objects near catcher are difficult to detect, Kalman filter is applied to estimate moving trajectory of baseball. Finally, ball trajectory of three-dimension is reconstructed and database of video stream is generated automatically. The experiment result shows that the system builds 2D and 3D trajectory successfully. In the scenario of small space the difference between the measured values of strike zone border and actual values of border are 3.01%, 6.95% and 1.79% respectively. The average difference is 4.32%. The average capture time of baseball trajectories is 50ms. This nearly meets real-time.