This paper presents a method for\nsynthesizing a stroboscopic image of a moving sports\nplayer from a hand-held camera sequence. This method\nhas three steps: synthesis of background image,\nsynthesis of stroboscopic image, and removal of player�s\nshadow. In synthesis of background image step, all\ninput frames masked a bounding box of the player are\nstitched together to generate a background image. The\nplayer is extracted by an HOG-based people detector.\nIn synthesis of stroboscopic image step, the background\nimage, the input frame, and a mask of the player\nsynthesize a stroboscopic image. In removal of shadow\nstep, we remove the player�s shadow which negatively\naffects an analysis by using mean-shift. In our previous\nwork, synthesis of background image has been timeconsuming.\nIn this paper, by using the bounding box\nof the player detected by HOG and by subtracting the\nimages for synthesizing a mask, computational speed\nand accuracy can be improved. These have contributed\ngreatly to the improvement from the previous method.\nThese are main improvements and novelty points from\nour previous method. In experiments, we confirmed\nthe effectiveness of the proposed method, measured\nthe player�s speed and stride length, and made a\nfootprint image. The image sequence was captured\nunder a simple condition that no other people were in\nthe background and the person controlling the video\ncamera was standing still, such like a motion parallax\nwas not occurred. In addition, we applied the synthesis\nmethod to various scenes to confirm its versatility.
Loading....