Automatic analysis of complex athlete techniques in broadcast taekwondo video

TUHAD: Taekwondo Unit Technique Human Action Dataset with Key Frame-Based CNN Action Recognition

“…The performance of the proposed key frame-based taekwondo action recognition algorithm was quantitatively compared with the previous algorithms [ 39 , 40 ] applied for taekwondo action recognition regarding the recognition accuracy and computation time required for the action recognition. Since the previous algorithms were specialized for their own datasets, it was difficult to apply the algorithms to TUHAD.…”

“…The recognition algorithm proposed by Seo at al. [ 39 ] has a simple structure that calculates the image histogram and classifies it. However, due to the limitations of the method using simple histogram, it seems that the accuracy decreases in the dataset with diversity of brightness and subjects.…”

“…Choi et al [ 38 ] proposed a system for remotely scoring poomsae by comparing actions acquired from multiple vision sensors with corresponding reference actions. Seo et al [ 39 ] suggested a Poisson distribution-based recognition algorithm that uses one-dimensional spatial information extracted from image sequences as its input. Kong et al [ 40 ] extracted taekwondo actions from a broadcast competition video and classified them with high accuracy by applying a support vector machine and defining a taekwondo action as a set of poses.…”

“…Kong et al [ 40 ] extracted taekwondo actions from a broadcast competition video and classified them with high accuracy by applying a support vector machine and defining a taekwondo action as a set of poses. However, these studies had limitations such as low accuracy despite a complex recognition system [ 38 ], restricted applicability arising from limited movements [ 28 ], and vulnerability to the movement of subjects because of the exclusive use of histogram images [ 39 ]. Despite a high recognition accuracy rate, Kong et al’s method [ 40 ] requires two training and recognition models each for a single frame and action classification process, and the pose order needs to be defined for all action classes.…”

Automatic detection technology of sports athletes based on image recognition technology

“…Target segmentation is a key step in the identification process, which usually means separating the target we are interested in and the background we are in as an independent part [1]. Traditional segmentation methods include background subtraction, interframe difference and optical flow, as well as skin color detection for human detection.…”