Toast CAM, an integrated smart store management system that provides all services necessary for store operation, uses self-developed cameras for facility security, customer congestion analysis, and face recognition for entry/exit controls.
NHN ToastCAM is seeking advancement of the face tracking modules for their existing system by applying features that extracts a representative image from multiple recognized images as well as consulting in the field of vision AI to minimize cost of cloud servers due to re-identification process.
Modelling & Test set generation
Development of face recognition(detection) module
Vision AI Consulting
System integration & technology transfer
Technological Challenges
Advancement of Face recognition and tracking module
To accurately track the movement of the person after facial recognition, repetitive inquiries of the recognized face are performed through communication with the cloud servers, which leads to an increase in operation cost. Therefore, system should recognize a face more quicky, extract only one representative face from multiple faces tracked, and able to specify time duration for inquiries to minimize communications with the server.
1
Module performance testing and integration
The module should achieve the same face tracking error rate 10% or less, and the ratio of the number of representative images extracted should be adjustable according to the number of pictures taken consecutively. Exclude overlapping and high-speed movement, the module should receive image sizes and formats required by AI system without affecting the performance of ToastCAM to generate results.
2
Road Map
Development of face recognition(detection) module
Development module test/improvement
Consulting
Support education/technology transfer
Key Features
Face identification advanced module
1
Operation within AI System settings
(Image sizes, formats/Representative face selection cycle/Face Identification reset cycle)
2
The Result
under
4FPS
Face Identification Speed
90%
above
Face Identification Accuracy
100%
Integration with existing systems
