Shivani Pachika

 

Date:  Friday September 27, 2019

Time:  1:00pm-3:00pm

Location: 3105, Lambton Tower

 

The rapid advancements in artificial intelligence have enabled recent progress of self-driving vehicles. However, the dependence on 3D object models and their annotations collected and owned by individual companies has become a major problem for the development of new algorithms. This thesis proposes an approach of directly using graphics models created from open-source datasets as the virtual representation of real-world objects. This approach uses Machine Learning techniques to extract 3D feature points and to create annotations from graphics models for the recognition of dynamic objects, such as cars, and for the verification of stationary and variable objects, such as buildings and trees. Moreover, it generates heat maps for the elimination of stationary/variable objects in real-time images before working on the recognition of dynamic objects. The proposed approach helps to bridge the gap between the virtual and physical worlds and to facilitate the development of new algorithms for self-driving vehicles.

 

Thesis Committee: