We developed a solution to detect and track athletes in marathon videos using face and bib number recognition. The system runs on a simple intel i7 machine without a GPU and achieves near real-time performance at 7 fps. Our AI engine resulted in a 97% detection and tracking accuracy, leading to the creation of personalized video clips for registered participants. The client’s productivity increased by 300% with this solution. The technology stack includes Python, Yolo, OpenCV, and deep learning models.

Business Challenge:
The client installs his cameras at various location of the marathon track and records video. A few participants registers with the client. The use case is to detect and track automatically the registered athletes from the marathon’s complete race videos and clip out personalized short videos at each spot the camera is installed.

Solution:
We developed and optimized a solution that performed the automatic detection and tracking of athletes by face and their bib number both available on front of the shirt. The system is deployed on a simple intel i7 machine with out GPU that performs the said task in near real time at 7 fps.

AI Engine Results:
We achieved 97% detection and tracking results to clip out small videos of athletes.

Results:
An app that churns out the personalized video clips of the registered participants in the marathon.

Client Success:
The automatic personalized video extraction increased the productivity of the client by 300%.

Technology Stack:
Language: Python
Technologies: Yolo, OpenCV, Deep Learning Models

Photo Courtesy: olympics.com

We developed a sports analytics solution that works with both Android and iOS mobile devices. Our approach uses a YOLO and DeepSort-based model tuned on Tensorflow-lite to conduct real-time pose estimation of players and ball identification and tracking in baseball and golf. Our AI engine results in two Tensorflow lite models running on high-end Android and IOS devices. Without employing pricey hardware, the client was able to launch a first-of-its-kind, cost-effective sports analytics tool. The project’s technology stack consists of Yolo, DeepSort, Deep Learning Models, and OpenCV.

Business Challenge:
Every sport has introduced the field of Analytics. We worked on the analytics of two sports, Baseball and Golf. There were three major challenges. The first challenge was to detect the pose of the player. The second challenge was to detect and track the ball moving at around 160 km/h in near real-time. The third challenge was to implement the solution on both Android and IOS devices.

Solution:
We developed and optimized a solution that performed the pose estimation of the player. Furthermore, we optimized and implemented a YOLO and Deepsort-based model that detects and tracks the ball. Finally, the model was optimized on Tensorflow-lite to run on Android and IOS devices.

AI Engine Results:
We developed two Tensorflow lite models that runs on high-end Android and IOS devices. The first model detects and tracks of the ball in near real time, whereas the second model detects the player’s pose .

Results:
An app that performs the pose estimation, and ball detection and tracking in real time.

Client Success:
The client was able to introduce a first of its kind mobile app that performs a reasonable analytics of the two sports like baseball and Golf without using specialized and expensive hardware that may cost as high as $20000.

Technology Stack:
Language: Python
Technologies: Yolo, DeepSort, Deep Learning Models, OpenCV

Photo Courtesy: Tech Bullion