International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 06 | Jun 2024
www.irjet.net
p-ISSN: 2395-0072
Automatic Key-Event Extraction from Sports Videos Using Scoreboard Detection 1SINDHUJA K, 2EYAMINI C, 3BRINDHA S 1Assistant Professor, Department of Computer Science and Engineering, EASA College of Engineering and
Technology, Coimbatore, Tamil Nadu, India
2Assistant Professor, Department of Artificial Intelligence and Data Science, Kathir College of Engineering,
Coimbatore, Tamil Nadu, India
3Assistant Professor, Department of Computer Science and Engineering, EASA College of Engineering and
Technology, Coimbatore, Tamil Nadu, India ---------------------------------------------------------------------***--------------------------------------------------------------------memory) [5] and [6]. Shingrakhia et al. (network). J.Yu et al. Abstract: This paper introduces an efficient method for
The Entertainment Industry and Sports Videos Sports play a significant role in the entertainment industry, attracting a vast audience worldwide. With the increasing accessibility of the internet, sports viewership has surged. However, many individuals struggle to find time to watch full-length matches due to their busy lives. As a result, there is a growing preference for shorter, summarized versions of sports events. Manual summarization or highlight generation is a labor-intensive process that requires professional video editing tools. Unfortunately, this approach imposes limitations on the amount of video footage that can be effectively summarized within a given timeframe.
Although these are new techniques, they all focus on a single aspect for the subject and are difficult to translate to other movements. Others, such as P. Kathirvel and others, have attempted to generalize across different sports. This works for games that don't require a whistle or for background noise etc. It fails when the whistle cannot be heard due to the neural network works and uses it to identify important events in the game called Kendo (Japanese fencing). This includes cricket, football etc. It doesn't work very well in multiplayer games like. In this paper, we propose a computationally cheap method to extract outcome-based values from a scoreboard about a game. Call YOLO and train it to determine the desired score of the game. We then run a sample of images taken from the movie at a frame rate of one frame per second and run OCR on them to keep track of the scores. If the score is changed by some forward increment (for example, in a baseball match), the increment of the variable score will be set to 4/6 and the increment of the wicket difference will be set to 1 because the difference between the wicket1 indicates that it is an event and therefore should occur in the context of the match. In football, the delta will appear during the match and the increment will be closed correctly. Our model is very simple, only need 250-300 points of footage to train and anyone can use it for any sports video using the method with knowledge of sports to adjust the time increment. The remainder of this article is organized as follows. Section 3 describes YOLO and why we chose YOLO, Section 4 presents the methodology, Section 5 presents the results and performance of our model, and finally Section 6 presents conclusions and limitations/future work.
2. LITERATURE
2.1 YOLO (You Only Look Once)
Some of the previous work in this area includes the use of ORB (Oriented Fast, Rotated Brief) to identify BRS (bowler run sequence) in cricket and classify them as important [1]. Pushkar Shukla et al. In [3], they used the intensity of the operator's gestures to determine the occurrence of events. A. Javed et al. wicket). Extracting features from football videos for content using 3D CNN (3D convolutional neural network) and LSTM (long-term
2.1.1 COMPARISON OF OBJECT DETECTION ALGORITHMS
extracting key events from sports videos by leveraging scoreboard detection. The approach involves training a supervised learning-based object detection algorithm (YOLO) on a dataset of 1200 images. Once the scoreboard is detected in video frames, it is cropped out, and image processing techniques are applied to reduce noise. An Optical Character Recognizer (OCR) extracts the score, and a rule-based algorithm generates precise timestamps for critical moments during the game. The proposed method achieves an average F1 Score of 0.979 across different sports, making it valuable for sports analysis. This implementation is in Python 3.7. Index Terms - Automatic Key-Event Extraction, Sports Videos, Scoreboard Detection, YOLO (You Only Look Once), Image Processing, Optical Character Recognizer (OCR), Timestamp Generation.
1.INTRODUCTION
© 2024, IRJET
|
Impact Factor value: 8.226
There are many target detection algorithms such as R-CNN [12], Faster R-CNN [13], and YOLO. In this section, we will discuss the reasons why we prefer YOLO. This is done by showing other algorithms, their limitations, and how YOLO addresses/overcomes these limitations. Region based convolution Neural Network (R-CNN): The aim is to divide
|
ISO 9001:2008 Certified Journal
|
Page 1015