Technical Name Deep-learning-based object recognition, behavior,360-degree video SLAM technology for autonomous driving
Project Operator National Chiao Tung University
Project Host 郭峻因
Summary
ezLabel features automatic route predictionfitting algorithm, which reduces the time to labelensure the quality , collects  various sample,help to customize AI function. Rear vehicle overtaking prediction uses C3D-based deep learning network with 16 rear camera images. It can be applied in E-mirror products to ensure safer driving. Besides, a 360-degree video SLAM technology solves the drawback of finite FOV video SLAM, achieves better accuracy,speeds up 3D map establishment.
Scientific Breakthrough
"ezLabel:
1. Route prediction: label object with 2 frames
2. Fitting: speed upguarantee data quality.

Rear vehicle overtaking prediction:
1. Heat-map shows overtaking
2. 3D CNN implementation
3. Achieve 95 accuracy rate at daynight
4. Detect objectrecognize behavior at same time.

360-degree video SLAM:
1.	Supporting different amount of cameras
2.	Positioning with panoramic i
Industrial Applicability
ezLable gains the award of Audi Innovation Award. ezLabel as the basis of developing deep learning-based video function tool is able to speed upguarantee the quality of labeled data.These AI  systems can be applied in to industry 4.0 for AGVstore warehousing robot, hospital for the patients, the disabilities,the weakness, airport for luggage  transfer,autonomous car for locating.
Keyword __
Notes
  • Contact
other people also saw