Object Detection and Terrain Mapping in Industrial Scenes

Computer Vision techniques and algorithms, especially those involving deep neural networks, are now predominant in robotics and autonomous driving. These algorithms have found immense success in urban scenes for autonomous driving, and visualize 3D maps of cities. However, like most deep learning solutions, they do not generalize well over a large varieties of inputs. Thus, while urban scenes are well represented in computer vision research, less common scenes, such as construction sites, mines, quarries, and farms remain largely unexplored. Algorithms in obstacle detection to moving
vehicles in hazardous worksites can prevent accidents which often cause human injury and damage to properties. Additionally, building maps of such worksites can present a whole host of opportunities in worksite surveillance, safety, and autonomous driving in such landscapes. Lastly, industrial worksites tend to be in a constant state of flux, unlike many urban environments. This gives us an opportunity to develop algorithms capable of withstanding such changes, thereby contributing valuable research to the field. This project is bifurcated into two parts, one part that focuses on 3D automated
object detection and avoidance on industrial worksites, and the other part on building a 3D map of the worksite for field inspection purposes.

Faculty Supervisor:

Nilanjan Ray

Student:

Partner:

Correct-AI Inc.

Discipline:

Computer science

Sector:

Manufacturing

University:

University of Alberta

Program: