Context Based Multiple Players Identification in Sports Images
In this paper we propose to develop technology for recognizing Jersey number of each athlete in real time to do sports video indexing in multi-player game. More than two persons are often detected as one object due to frequent collisions between persons i
- PDF / 1,201,264 Bytes
- 5 Pages / 439.37 x 666.142 pts Page_size
- 70 Downloads / 161 Views
Abstract. In this paper we propose to develop technology for recognizing Jersey number of each athlete in real time to do sports video indexing in multi-player game. More than two persons are often detected as one object due to frequent collisions between persons in sports. This system recognizes the jersey number of more than one persons based on learnt context information using texts and uniform color information. The resulting system is able to achieve a jersey number recognition accuracy up to 97% on video and photography. Keywords: Jersey number recognition
Color context SVM
1 Introduction Sports video indexing technology has been researched in various meaning levels. Mainly, there are shot level [1], event level [2], and action level [3]. Recognizing individual player is a very important technology in these various video indexing technologies, and it is used to provide various real time broadcasting services if various broadcasting information are combined [4–8]. This paper is to develop the technology which recognizes Jersey number in real time in the detection area of actually playing athlete and does video tagging by relating this information with existing registered athlete. The athlete is very fast which makes adjacent or overlapping conditions occur frequently. This paper makes it possible to discriminate multiple objects and recognize the Jersey number accurately by extracting uniform and Jersey number in the area of complex objects after learning representative color of uniform and Jersey number.
2 Context Based Jersey Number Recognition Algorithm The proposing team and Jersey number recognition system is as Fig. 1. It is composed of algorithm which recognizes team by using color information, module which detects Jersey number area, and algorithm which recognizes Jersey number based on Jersey number detection results.
© Springer Nature Singapore Pte Ltd. 2017 J.J. (Jong Hyuk) Park et al. (eds.), Advanced Multimedia and Ubiquitous Engineering, Lecture Notes in Electrical Engineering 448, DOI 10.1007/978-981-10-5041-1_114
716
M. Nam et al.
Fig. 1. Team recognition/Jersey team number recognition system architecture
2.1
Color Based Team Recognition Algorithm
In sports indexing system, tagging each athlete’s team information automatically is one of the very important functions for various broadcasting services. We develop algorithm for identifying teams automatically using quantized color information. Figure 3 shows the algorithm for team identification. Team identification is conducted by learning and recognition process. Learning is the process of extracting representative colors for each team. The recognition is based on color value threshold using representative color of each team. The figure below shows the quantized RBG image of input image.
2.2
Context Color Learning Algorithm for Multiple
The learning module for classifying and recognizing Jersey number using context color information in the area of complex objects is as follows: (Quantize and save team Jersey number and Jersey RGB col
Data Loading...
