The use of advanced CV (or “Computer Vision” applications) in sports ultimately allows for a highly efficient, fast, and precise analysis of actions, conditions, and environments in all possible sports events. The naked human eye is gradually being replaced with smart algorithms that do all the cumbersome analytics automatically. These capacities may help better analyze the crucial sports event moments, allowing us to receive more precise scores and judge more efficiently as a whole.
Although the use of CV in professional sports mostly requires pre-recorded content of high-definition video, the technology is pretty in-depth and efficient in processing video of any format from any device. With this in mind, it’s interesting to wonder where this form of machine processing might best be applied. Let’s dive a bit deeper into the topic of using content processing via computer vision in the sports industry.
Common Moments Efficiently Analyzed via CV
Besides used in security systems where image quality is crucial (for the recognition of faces, dangerous objects, etc.), machine vision technologies are used in many other cases in sports, including:
- The training process — the in-depth analysis of action that CV can provide makes it incredibly helpful during training sessions where athletes can examine their technique through video.
- Refereeing —CV allows for three-dimensional simulations and video inspection of the offsides, outs, goals, photo finish in mass races, and more.
- Rates of player activity during events — for instance, in tennis the dynamics of players can be captured and analyzed based on their movements and even slight gesture on the court.
- Ball (puck, dart, shotput, you name it) trajectories — these can be analyzed and predicted for further in-depth analytics.
- Action camera stabilization and smart focus — artificial intelligence in sports allows for the real-time smoothing out of action frames and automated focus based on the density of activity and target actions in the field.
These are just some general capabilities that modern sports events organizers get to employ thanks to CV. Let’s take a look at a bunch of particular cases in greater detail.
Computer Vision Applications in Sports
Real-time action recognition in hockey
Specialists at the Shiraz University and the University of Waterloo dedicated an entire paper to efficient recognition of actions in hockey. The main gist of this paper is that experts have come up with the so-called “Action Recognition Hourglass Network (ARHN)”, which is a complex, multi-component visual data processing model.
In simple terms, the complex algorithm takes a piece of motion video content and converts it into a series of images. Another underlying algorithm within a stacked hourglass network then analyzes players’ positions (straight and crossover skating, pre-and post-shot poses) and classifies them.
These models have been used for the longest time to help issue the fairest, most precise scores in hockey and other types of sports out there.
Ball tracking systems in tennis
Precise tennis (as well as badminton and cricket) ball trajectories have been tracked in sports since the mid-2000s. Thus, specialized systems focus on multiple objects in the image that are similar in form to a ball. Upon detecting these, a three-dimensional trajectory is built by connecting the ball movement pattern frame by frame.
This is where multiple camera angles and flexible motion capture are essential. The main purpose here is the precise statement on whether the ball landed in or out of bounds during the game. On their deepest, most complex layers, the underlying algorithms can build predictions of ball trajectories based on various conditions (a player’s miss, for example).
Based on such solutions, smart statistics are generated in real-time for 100% fair refereeing and reputable sports performance analytics.
Training activities analytics
Modern sport imposes higher demands, not only on the athletes but also on the team of coaches. The key advantage in team sports is not so much the presence of “stars” as the proper organization of the team game, the assessment of the actions of each player, their interaction. Because of this, it is crucial that the coach develops effective tactics and game strategies.
Computer vision in sports analytics is a great tool for getting objective, up-to-date information in the conditions when solely recording a video of a game field is not enough. Mathematical processing of video streams allows for the analysis of each player’s position throughout the game.
For many sports arenas and clubs, sports video analytics systems have now become a very profitable business. Even though the creation of such systems requires organizing the synchronous operation of dozens of cameras and powerful computing capabilities, the effort usually pays off handsomely in the long run.
Prevention of life-threatening situations
In NASCAR racing and similar kinds of sports where players experience potential life dangers, computer vision is used to detect and prevent vehicle malfunctions in a timely manner. This is where such systems can save lives.
Commonly, large big data-based databases of vehicle models are implemented to recognize particular cars, analyzing them in the tiniest detail during the event. In this way, experts get real-time reach into the car’s stats to track any small malfunctions which could lead to serious consequences.
Fan mood and engagement analysis
A not so obvious application of machine learning in sports analytics is that it provides organizers with the ability to recognize faces on the tribunes and identify emotions that fans experience during the sporting event. This is meant to stimulate the hype on the tribunes and build statistics on fan engagement as well as an event’s overall impact.
Smart sports journalism
Adding on to the previous point, and expanding on the influence of a sports event on fans, computer vision can also be beneficially used to generate impressive media content and more precisely report on the game highlights. By analyzing the most outstanding, dynamic actions happening in the field (or track, ring, court, etc.) using the above-mentioned algorithms, it’s possible for journalists to focus on the most exciting occurrences in that environment.
This is a crucial capability when it comes to live events as it helps to keep all spectators on the edge of their seats. Apart from visual features, AI even helps to automatically commentate events without the aid of live speakers (automated insights, for instance, developed a solution for real-time narratives based on “Natural Language Recognition” capacities).
The Specifics of Software for Computer Vision in Sports
The above applications and more make the world of sports a firmly-watched, ever so exciting, and competitive realm to organize. There are various types of solutions in the niche. Some of the leading examples include Sentio’s smart tracking and analytics systems; Stats Perform’s SportVU 2.0 with in-depth computer vision-based algorithms; GAMEFACE.AI with its in-depth analysis of strategic insights and other footage points; and more.
The available solutions are intricate systems to be integrated through hardware and software by a dedicated integrator specialist. The role of the integrator is limited to adapting the system based on the readymade standard components according to the requirements of a particular customer, its binding to a specific object, installation, and service entry. Thus, the resolution and speed of the cameras are limited by the capabilities of the human operator, and the main focus is made on minimizing the volume of video recordings and the convenience of working with them.
Crucial points for getting the highest-quality analytics
The industry of computer vision system for tracking players in sports games has slightly different priorities arising from a much wider range of tasks, causing a very limited distribution and use of “boxed” products. Due to the diversity of objects and tasks for observation, the requirements for image capture systems vary greatly.
First of all, machine image processing is supposed to entail requirements for the maximum transfer of details, diversity, and uniformity of shooting conditions to increase the efficiency (showing details), speed, and reliability (shooting conditions) of the algorithms. Based on our team’s experience, the main points in the selection of machine vision components are as follows:
- Image quality, degree of detail, and speed (frame rate) must correspond to the mathematical algorithms used to solve various applied tasks.
- Lighting conditions should be as stable and/or controlled as much as possible. In most cases, artificial lighting is used.
- Limited use or complete absence of automated functions such as auto exposure or autofocus in the camera. Everything is controlled by external software.
- The main information processing is performed on separate calculators since the complexity of the algorithms does not allow placing the required computing power in a compact camera body. In some cases, joint processing of images from multiple cameras is required. The type and power of the calculator are determined by the requirements of the specific task and the math used.
- High-speed interfaces for transmitting images with high resolution (details) and high frame rate (fixing fast processes) are required.
- Software functionality from camera manufacturers is limited to a set of drivers for the flexible configuration of equipment. We develop application programs for each specific project.
Our Computer Vision Use Cases
Recruiting analytics platform for tracking the speed of football players. Design and development by Requestum
Our team has experience in using computer vision in American football. The system was mainly used in recruiting. The customer requirements for functionality included:
- The ability to track player speed,
- Tracking and monitoring the effectiveness of players,
- Visual information for further analysis of game situations with the team.
As a result, when selecting athletes, a recruiter doesn’t have to attend competitions or travel to assess their abilities. They just have to watch a video with all the necessary analytics in the office or at home.
Conclusion
Artificial intelligence in sports makes refereeing, analyzing, highlighting, and satisfying fans easier to grasp and more efficient in the long run. When it comes to implementing an AI-based system for sports events, you have the ultimate choice of going for renowned yet costly solutions or ordering a cost-efficient custom local system.
Originally published at https://requestum.com.
