Challenges for Person Re-Identification
What do you think would happen if we lived in a society where every action you took, where you went and what you were doing was tracked by a camera?
VFA’s Mobile AI Lab team is researching about Person Re-Identification to recognize a person by associating different images of the same person taken from different cameras or from the same camera in different time periods.
From a variety of different images of people in different poses (walking, running, standing, sitting, etc…), it detects everybody appearing in video and attaches an ID for them after querying the data warehouse. It is very useful for a number of public security applications such as intelligent camera surveillance systems or anti-theft cameras. Recent scientific studies show that the accuracy of person re-identification is still not high enough to be used in real life.
This Person Re-Identification is different from MAL Face Recognition (our product) which recognizes a registered face embedded in the kit. A list of cropped images of people is retrieved as candidates and will contain the desired people possessing the same identity to the search target. The 3 main types of difficulties we are encountering:
- Same person at the different situation
- Same person in different clothes
- Low-resolution image
In the future, the main applications of person re-identification are expected to be used for systems of security, marketing and so forth in commercial facilities using monitoring cameras installed outside.
It could track would-be criminals. Marketers could analyze people's purchasing behavior.
We would be tracked everywhere, 24/7.
