1. Person Identification
   1. algorithms
      1. EfficientDet
      2. Yolov5
      3. Retina Net
   2. datasets
      1. Crowd Human
2. Deep Person Re-identification
   1. algorithms
      1. Strong ReID baseline
      2. ResNet50
   2. datasets
      1. Market 1501
      2. Peta
3. Person Tracking
   1. Deep SORT
4. Face Detection
   1. RetinaFace
   2. MobileNet 0.25
5. Face Re-identification
   1. ArcFace
6. Age and Gender Prediction
   1. efficientnet_b0
   2. VGG
7. Face mask detection
   1. use key points from RetinaFace
      1. we can compare the color in the area of ​​the eyes and in the corners of the mouth, if they are approximately the same, then there is no mask
8. Search for abandoned items
   1. dynamic background subtraction SubtractorMOG2
      1. We do not take person into account in the background, we remove them using the person detector
      2. Additionally, we remove all small objects using morphology
9. Smoke/ fire detection
   1. ResNet50 classifier into three classes - normal / fire / smoke
   2. Divide the image into ~ 10 sections and run the classifier on them
      1. if at least one fire / smoke is detected for several frames, then we give an alarm
10. Sabotage detection
    1. ORB
       1. We dynamically remember the ORB descriptors of the image, if they have changed a lot, it means there was camera sabotage