Research And Implementation Of Real-time Multi-object Localization And Resetting Method

With the development of sport, more and more science technology is needed insport training and competitions. Video based sport training and competitionssupporting system could be applied in a lot of application scenarios. Nine ball is oneof popular indoor sports. It is also one of famous international professionalcompetitions.In this thesis, we study the problems of multi-object location, so that the locationof each ball could be determined. Furthermore, a resetting method is proposed, so thatthe objective training could be implemented. And the capability of the player could beraised in the limited time. The details are as follows:1. We propose a simplified background modeling algorithm, which could modelthe background in Gaussian function. And the value of Gaussian function isrepresented in a look-up table. In the stage of object detection, the backgroundsubtraction could be obtained by look-up table, so that the running speed is reduced.2. We propose a multi-image multi-threshold object detection method. Byextending the color space, we obtain the complementary images, so that the distancebetween object ball and background is increased. Three level Edge detection isimplemented in the multi-color spaces. Finally, we detect the circle center of all theballs using improved Hough Translation based on6thresholds.3. We design color and edge statistics based object recognition algorithm. Thecolor statistics in quantified HIS space of all balls is obtained from the training set.For most of balls, the ball number is determined by the color statistics. The ball ofnumber1and number9have the same color of yellow. The edge statistics is furtherextracted to distinct these two balls.4. We proposed a reset method based color and shape features. The searchingscheme is improved and the searching area is shrinked, so that the reset accuracycould be protected and the computing time is reduced.