AI DEVELOPMENT CASE STUDY

Paracel company has developed many types of AI such as: Indoor facial recognition, Automatic house drawing design, Number plate recognition…

In-house face recognition

  • SubjectThe company has a staff of more than 100 people and has many visitors every day. The system must recognize staff and customers with high accuracy.
  • PTS’ Solution:
    • We implemented advanced solutions for face recognition systems and research results on face recognition systems, and solved the problem by learning only a few captured images.
    • Using UltraLite for face detection reduces search times to less than 10 nano seconds. Faces are aligned using dlib, face landmarks are detected and tracked by the SORT algorithm.
    • We used a few shots of MobileFaceNetv2 trained on the Siamese Network to reduce the number of parameters and run them in real time on the CPU without compromising accuracy. Using the provided staff images, we incorporated and trained a machine learning model using metric learning to classify people.
Aiface
  • Result
    • Using a real-time camera, with a few learning sessions, we achieved high accuracy using only a few images of each staff member.
    • The accuracy is 99.3% when running the application on a recognized person with a front photo.
    • The detection accuracy of unknown people (people not included in the database) is 98.5%.
    • Face tracking and calculation with 99% accuracy are supported for multiple faces.
    • It supports both Asians and Westerners.

Housing Auto Design

Input: An image of geometry of a land.Output: An architecture design for a house on this land.

Diseased pig detection

animal tracking

Input: Streaming video.Output :The frame in which there are diseased pigs sitting like a dog.

Vehicles detection

Vehicles tracking

Input: Streaming video.Output :The frame in which there are detected vehicles.

License plate recognition

License tracking

Input: Streaming video.Output :The frame in which there are detected and recognized plate.

Face emotion

face tracking

Input: The frame in which there are facial emotions.Output: The frame in which there are detected emotions on those faces.

feel tracking
feel tracking

Human gesture project

Human gesture tracking

Input: Streaming video.Output: The frame in which there are detected humans skeleton.

Heatmap

heatmap

Input: Streaming video.Output: The frame in which there are detected and counted people with heatmap.