A judgment framework for roller skating using faster R-CNN

Intelligent sports game referee is a new research direction in image processing and deep learning field. In this paper, we present an image processing framework for judgment of the speed slalom race in roller skating competition. Firstly, we chose faster R-CNN algorithm to identify targets such as pile and person during the race. Then an image processing framework is designed to process the recognized target in order to get the competition result. In this process, we remove the duplicate boxes after object recognition, modify the scale in RPN according to the target size and eliminate illumination influence by edge detection. After this, we complete our framework and experimental results show that the proposed framework is efficient and effective in terms of our database.

I. Introduction

With the development of image processing and deep learning technology, more and more practical problems can be looked at from the perspective of image processing and solved by deep learning algorithms. In the field of sports, especially in the field of SLALOM, we can also use image processing and deep learning methods to analyze the problem. For example, in the roller skating competition, we can finish the roller skating referee's work by processing the video of the speed slalom race. In the skiing competition, a skier must go through a passage between two upright poles in a slalom race. We can decide the result by image processing.