Facial recognition and tracking
- #Computer Vision
The DataBase of the faces with time report from the video.
- Data Analysis
- Machine Learning
Impact
The solution helps to understand who was present during the video, when, and for how long.
Services we provided
The database of the faces with time report from the video
Tech Stack
Pytorch
Huggingface
RetinaNet
DeepFace
Challenges and Solutions
🧐 Challenges
- To develop a pipeline that incorporates three different technologies: face recognition, face detection, and face tracking.
- To find the best models and practices that can handle different lighting, image resolution, and occlusions.
💡 Solutions
Our solution successfully implements the processing pipeline, db creation, and report forming. The processing pipeline includes the following steps:
- Extracting faces with a CNN (RetinaFace) every xth frame and passing results to the face tracker.
- Tracking faces until we have a new id.
- Recognize the new face, and tries to find out if the person was encountered before or not.
- Adding new information to the db and report.
User flow
1. The user uploads the video, and several arguments, and starts the processing.
2. Pipeline gathers the information, creates the db, and the report.