Abstract:Nowadays, it becomes a common practice to capture some data of sports games with devices such as GPS sensors and cameras and then use the data to perform various analyses on sports games, including tactics discovery, similar game retrieval, performance study, etc. While this practice has been conducted to many sports such as basketball and soccer, it remains largely unexplored on the billiards sports, which is mainly due to the lack of publicly available datasets. Motivated by this, we collect a dataset of billiards sports, which includes the layouts (i.e., locations) of billiards balls after performing break shots, called break shot layouts, the traces of the balls as a result of strikes (in the form of trajectories), and detailed statistics and performance indicators. We then study and develop techniques for three tasks on the collected dataset, including (1) prediction and (2) generation on the layouts data, and (3) similar billiards layout retrieval on the layouts data, which can serve different users such as coaches, players and fans. We conduct extensive experiments on the collected dataset and the results show that our methods perform effectively and efficiently.
What problem does this paper attempt to address?
This paper aims to solve several key problems in billiards motion data analysis, which have not been fully explored mainly due to the lack of publicly available data sets. Specifically, the objectives of the paper can be summarized as follows:
1. **Data collection and collation**:
- A data set containing billiards match layouts, trajectories, and detailed statistical information has been collected. This data set covers 3,019 match records, 6,637 round records, and 2,082 shot records from 94 international professional 9 - ball matches in the past 20 years.
2. **Layout prediction**:
- **Task 1: Opening layout prediction**: Given a match layout, predict the results in three aspects:
- Whether to clear the table: Predict whether the player who executes the opening shot can pocket all the balls.
- Whether to win: Predict whether the player who executes the opening shot can win the match.
- The number of consecutive balls pocketed: Predict the number of balls pocketed consecutively after the opening shot.
- Use a convolutional neural network (CNN) model (called BLCNN) to complete these prediction tasks. BLCNN is trained by supervised learning and can capture the spatial correlation between balls.
3. **Layout generation**:
- **Task 2: Opening layout generation**: Generate real and high - quality (i.e., easy to clear the table) opening layouts. This is crucial for players to understand how to deal with different layouts in actual matches.
- Use a generative adversarial network (GAN) model (called BLGAN) to generate these layouts. BLGAN generates a series of discrete tokens through the generator, each token representing a discrete position on the billiards table, and uses a discriminator to distinguish between real and generated layouts.
4. **Similar layout retrieval**:
- **Task 3: Similar layout retrieval**: Find billiards layouts similar to a given query layout from the database. This task has applications in multiple scenarios, such as real - time prediction of match results, coaches analyzing player performance, and fans looking for similar matches.
- Propose a method based on deep metric learning (called BL2Vec) for measuring the similarity between billiards layouts. BL2Vec not only takes into account the unique characteristics of billiards layouts but also can operate efficiently in linear time.
### Main contributions
1. **Data set contribution**:
- Provide a billiards motion data set containing opening shot data, shot statistical data, and trajectory data. These data sets support various machine - learning tasks and are publicly shared, providing a valuable resource for the research community.
2. **Prediction and generation tasks**:
- Study the prediction and generation tasks of opening layouts, and develop the BLCNN and BLGAN models. Experimental results show that BLCNN outperforms the baseline method in classification accuracy, while BLGAN performs well in generating high - quality and real layouts.
3. **Similar layout retrieval**:
- Propose a new problem, that is, retrieving billiards layouts similar to a given query layout from the layout database. Develop the BL2Vec method, which not only takes into account the unique characteristics of billiards layouts but also significantly outperforms existing methods in the similarity search task and has high operating efficiency.
### Related work
The paper also reviews research in related fields such as sports data analysis, sequence data prediction, generative models, point - set similarity measurement, sequence or structure similarity measurement, and deep metric learning, emphasizing the uniqueness and innovativeness of this research in billiards layout data processing.
In general, through tasks such as data collection, prediction, generation, and similarity retrieval, this paper comprehensively promotes the development of billiards motion data analysis and provides a powerful support tool for coaches, players, and fans.