Concentrate on the people who are not affordable the sustainable citizens like personal training
1) This gives the projects like the real time on device tracking of AR applications of single RGB camera.
Google의 Firebase (Backend service)
1) Authentication service
2) Real-time database service (NoSQL)
3) Cloud Functions (Rest API or Trigger function such as push function)
이 Firebase, database framework를 backend application으로 사용
Django web framework
MVC pattern 으로 part들을 쪼갠다 (Model, Controller, view의 개념)
- Django에서는 Controller가 아닌 Template으로 설명
User가 보이는걸 controller에 대한 것을 상호작용
- Model (Database)
MediaPipe framework
1) BlazePose detector
- detector
- removes the human region(사람이 있는 이미지 부분) from the input image
- estimator
- return key points from 256 * 256 image
-
Mobile Real-time Applications
Challenge
1) Hand gesture of both occluded and self-occluded hands (Wide range of hand size and large scale span)
2) BlazeFace detector
Implementation
- Health tracker + Web application using Django
- Exercise (Track of machine larning model)
Exercise Tracker
The location of 33 pose landmarks in MediaPipe Pose. (Using the BlazePose Detector)
Google Firebase , application by Google (Development platform)
- Webcam Use
- Dataset of 66k photos
- lower illuminiation -> reduce the accuracy
MobileNetV2-SSD
** MobileNetV2
1) CNN backbone to extract the feature from the input image (Made by Google)
2) CNN reduces the complexity and useless operation time
3) Mobile device에서도 좋은 성능 제공
** SSD
1) Bounding Box를 이용한 object detector로 YOLO와 같은 계열
Conclusion
(1) 기본 Mediapipe가 제공하는 BlazePose, BlazeFace vs (2) MobileNetV2-SSD
에서 (1)이 높은 효과를 보였다.
