Face recognition on sports training videos

Using face recognition on sports training videos, our client was able to reduce manual analytics efforts by more than 50% and get new insights on the team


Problem statement

Sports training institutions have a need to evaluate their players on their performance. This evaluation helps them to identify players strengths and weaknesses and give them appropriate training to build a competitive team. The client was a football training academy and they were in need of video management tools that could identify players on the training videos. They wanted us to build a system that can analyze the player performance and produce reports over a period of about the improvements of a player performance. To study the performance of the players, we were required to first identify the players and then search the video segments based on the player recognition. There was also a need to extract relevant player video segments for analyzing the metrics in depth.

Our solution

The central piece of the project is the development of a face recognition system suitable for a certain racial profile and age group. Since most of the face recognition software are biased towards the racial profile of the training dataset, the face recognition performance was very poor for our client’s use case. So we collected around 1000 faces and annotated 5 key facial points. We then trained a CNN to generate a 128 point face vector for every face. Using the 128 point face vector we trained a SVM based classifier for each player. The face recognition system comprises a player registration module wherein the system learns about each of the players facial features. The face recognition system is then able to identify a player on the video file using the pre-learnt facial vector. We packaged the face recognition module into a desktop application that is capable of ingesting long video files and splitting the video file into segments pertaining to each player. The user of the application can then search for specific segments on the video file pertaining to a specific player and perform analysis about the player.

Key metrics

This project was developed in a time frame of 12 weeks. With the tool that we developed, the coaches were able to quickly review training videos and focus more on the training rather than video management. On an average, each coach was able to save an average of 1 hour per day on the monotonous video management duties.

Technology stack

Logo for Dlib image processing library Logo for OpenCV image processing library

Key Metrics

We are a team that values efficiency, innovation, and the pursuit of excellence in everything we do. We are a high performance team that is passionate about bringing AI and Cloud computing technologies to a larger industry audience. We have accomplished so much in such a short amount of time!


Years in business


Happy clients


Completed projects

2 %

Hire top talent

100 %

Certified team

90 %

Client retentivity