Source: Deep Learning on Medium
How Deep Learning and AI is Changing the Sports Industry?
Deep learning technology has impacted almost every other industry. Sports is no different. Deep Learning is helping businesses associated with sports to expand.
From NHL to MLB, NBA, NFL, and NASCAR, almost every major sports league in U.S.A is now incorporating AI to expand its business.
According to Statista, the North American sports market is predicted to reach $80.3 billion in 2022. The revenue may come from merchandising, gate revenue, media rights, and sponsorships.
There are many different areas in the sports industry, where AI plays an important role. For example, wearable technology, marketing, automated journalism, and incorporating computer vision.
These smart Neural Networks can figure out various parameters affecting an athlete’s performance and provide recommendations to prevent an injury or fatigue.
These were only a few use cases that give an overview of the capabilities of AI and Deep Learning. Below mentioned are some of the top applications of Deep Learning in Sports.
1 — Sports Production
Deep Learning is used to create fully-automated sports production that looks almost identical to professional sports broadcast, including panning, camera zoom-ins on the action, etc.
The least we should expect from a decent-level automated sports production is to identify the players and ball. Identifying the ball is a difficult task in football. It can be anywhere on the ground, i.e., with players or in the hands of a goalie.
If you think about it, in all these distinct circumstances, the ball “looks” different, yet, we, as humans, have no problem distinguishing it as a ball from a single frame.
Recognizing the players is not straightforward either, as the system will have to differentiate between on-field players, bench players, and referees, etc.
2 — Recognizing the Court/Field
For the Deep Learning algorithm to figure out between ball and players, you first need to help it identify the field/court. Calibration will limit the scope of options for the Deep Learning algorithm by discovering within each frame which pixels are part of the court/field and which ones aren’t.
By determining the area of the court/field, it is possible to discriminate between players who are inside the court/field and others outside of it, such as coaches, bench players, spectators, etc.
3 — Data Annotation
As stated above, a part of the DL model training is a demand for an extensive data set to develop the “ultimate truth” for the DL algorithm. This is a key take away that should be done regularly as the algorithm is evolved, and more data is gathered.
There are many options to achieve this. Humans must annotate a minimal number of frames. Apart from these, several methods that require less effort are:
- YouTube/Google Images — It is possible to augment the data set by searching “football players” on YouTube or Google. This will produce images or frames that include football players; in other words, they have been “pre-annotated” as football players.
- Unsupervised Learning — This method uses un-labeled data by applying a supplementary “non-DL” algorithm to the first portion of the area of the potential players. For example, one can use known background subtractors such as MOG to recognize players roughly.
- Augmentations — Another usually used technique is to alter or augment the images. For example, modify angles, to stretch them, etc. These augmentations generate an additional data set that has been already labeled.
4 — Sports Marketing
Essentially every major sports league in the U.S.A could utilize additional marketing at any given point in time to increase that league’s resources and bolster their ROI.
Human intervention is crucial to make a marketing campaign successful, and DL algorithms can help sports workers achieve their goals.
Today, there are several human tasks can be replaced by AI and Big Data, including consumer research, to generate campaigns that help us to understand customer interests.
Content creation serves as another marketing tool where DL can play a part in. Moreover, technology can keep track of content performance to determine how well a campaign is doing.
As we’ve seen that with Deep Learning technologies, computers can comprehend sports action, opening new possibilities in sports production that were never plausible before.
If appropriately used, Deep Learning can mimic the decision-making process of a human camera operator as well as a video editor, providing the same experience of a professional live sports broadcast, at a reasonable cost.
This revolution in technology will allow semi-professional and amateur sports clubs to broadcast the games to their fans in a way that it can monetize their content.
Written by Sandeep Agarwal, CEO of Credencys a mobile and web app development company.