RoboComp Logo

A simple robotics framework.

GSoC’20 RoboComp project: Human recognition (identification) using multi-modal perception system

11th June 2020

Benchmarking methods for face analysis

Facial analysis is used to automatically label an individualin by retrieving it’s face features. It can be divided into 2 steps:

Detection

In each frame of the video we need to find the bounding box for each person’s face. To choose the technique to integrate, we compare their performance and computation time.

Top performers on UTK dataset

  1. Mt-CNN
  2. Haar Cascade

Face Detection performance

Face Detection time

Based on these results, we have choosen Mt-cnn as the face-detection model.

Recognition

Face recognition is a method of identifying or verifying the identity of an individual using their face.

Top performers on LFW dataset

  1. ArcFace
      2 versions of ArcFace are benchmarked. 
      - Using Resnet-100 as the backbone
      - Using MobileFacenet as the backbone 
    
  2. CosFace
  3. Face-net

Face Recognition performance

Face Recognition time

Based on these results, we have choosen Arcface as the face-recognition model.

Back To The Top


Current updates on the project

Please follow the following link for current evaluation of different methods of face recognition and gait recognition.


References

Deep Face Recognition: A Survey,
Mei Wang, Weihong Deng,
arXiv technical report (arXiv 1804.06655)

A survey on deep learning based face recognition,
Guodong Guo, Na Zhang,
Computer Vision and Image Understanding (DOI 10.1016)*


Shubh Maheshwari