Machine Learning, Computer Vision, and Drones at the Supermarket

07/03/2019 - 12:00
Speaker: 
Seminar: 
Where: 
Abstract: 

 

 

 

 

Machine Learning, Computer Vision, and Drones at the Supermarket

We will show that the following problems are surprisingly related, together with

provable approximations, and cool videos that demonstrate their real-world real-time performance.

1) Have a group of lawful (<200 grams) autonomous toy-drones that carry personalized ads inside the supermarket "Rami-Levi Hashikma LTD" near the university.

This is related to SLAM (simultaneous localization and mapping).

2) Given n lines and a n points, compute a rotation, translation and 1-to-1 matching that will minimize the sum of distances between every point to its matched line. 

This is a special case of a fundamental problem in computer vision, called PnP (Perspective-n-Point).

3) Minimize ||Ax-b||+\lambda ||x||-\lambda over every \lambda>0 and x in R^d. 

This is called Ridge regression in machine learning,  if lambda is given. We suggest the first provable approximation without tuning lambda .

4) Headache caused by current Augmented Reality glasses.

Partially based on papers in IEEE International Conference on Robotics and Automation (ICRA'19) and IEEE Robotics and Automation Letters (RA-L'19).

Joint work with Ibrahim Jubran, David Cohn, Ariel Hutterer, and Daniel Jeryes from the lab.