simple-attendance-system

Apply face regcognition to check attendance of student by webcam when they are in an online class. Up to now, I have developed 2 features:

• face recognition on one image which captured by user webcam
• face recognition on a video on user webcam.

#1. Tutorial: for face recogntion on a image

Step 1: download source from github

$ git clone https://github.com/nguyen-tho/face-recognition-with-deep-face.git

Step 2: install pakages by command.

$ pip install -r requirements.txt

Step 3: run program. run GUI.py file by command.

$ python GUI.py

or

$ python3 GUI.py

if you use python 3

#2. Collect data:

• System will take a video about 100 frames, user can make several poses as much as possible to create a variety dataset

#3. Capture an image to verify:

• When user need to check in, system will take a photo automatically for 5s after camera/webcam turned on #4. Deepface:
• In this project I have just used deepface to verify the identity of a person who sign up his/her information and data has been saved in our database
• Using verify method to check an image which is similar with one random image in his/her dataset images in database

from deepface import DeepFace
verified_img =  DeepFace.verify(image, image_in_dataset, enforce_detection=False)
#image is path of image which is captured to verify
#image_in_dataset is path of image which is in user's dataset
#output is a tuple contain verified status, facial area of 2 images, cosine similarity
#verified status: bool
#facial area [x,y,w,h] values
#cosine similarity is distance betwwen 2 vectors which embedded by 2 images. The less cosine the more similarity

• if verified value is True -> save log
• if not -> send alert "try again"
• In new version, I use find method to find identity of user who is owner of captured photo or video

  from deepface import DeepFace
  verified_list = DeepFace.find(image, dataset_path, enforce_detection=False)
  #image is path of image which is captured to verify
  #dataset_path is path to dataset folder
  #output is a dataframe of similar images list
  #each row has identity, facial area amd cosine
  #identity is path of image in dataset which deepface determine they are similar with captured image
  #facial area [x,y,w,h] values
  #cosine similarity is distance betwwen 2 vectors which embedded by 2 images. The less cosine the more similarity

#5. Advantages and Disadvantages

Advantages

• Can recognize object in weak brigtness environment
• When signed up user wear glasses. However, while recognition that person do not wear glass -> can recognize
• Have a good confidence (about more than 90%)
• DeepFace use pre-trained model -> do not need to train again

Disadvantages

• In the first time need to download model file and weights file -> it spends too much time (about 120 seconds) depends on computer and size if image database
• When add a new user, system will update new pkl weight file, it consumes too much time
• Sometimes recognition result may be wrong but confidence still high
• When compare with DeepFace.verify method, DeepFace.find method will slower than because need time to determine identity of the captured image
• Recognize ability maybe impacted by pre-trained model and detector backend.

#6. References:

• Deep face: https://github.com/serengil/deepface .
• UI design: https://github.com/joeVenner/FaceRecognition-GUI-APP

#7. New update:

• Solve the problem "cannot detect realtime" by current frame which taken by webcam and verify with user's image dataset
• Verify current frame with a random image in user's dataset
• Some detector backends and models can combine

  #some detector backend and model may suitable to use
  detector_backend = opencv, model_name = [VGG-Face, Facenet, Facenet512, ArcFace]
  detector_backend = ssd, model name = [VGG-Face, ArcFace]
  detector_backend = retinaface, model_name = VGG-Face # can use but slower than other detector backend
  #this is my personal knowledge about DeepFace. Can treat it as a reference

#8. Demo

Verify a person

Find identity of a person

9. Contact:

   Please contact with me at this email address to discuss: nguyencongtho116@gmail.com