Visual Semantic SLAM using Bottlenose Camera System

Simultaneous Localization and Mapping (SLAM) is useful in multiple applications such as autonomous driving, augmented reality, and surveillance. Visual SLAM is a popular and widely used SLAM technique due to simplicity of the sensor network. Semantics and machine learning are used widely in recent research to achieve better robust real time SLAM performance. Hardware acceleration can be used to perform repetitive and compute intensive parts of the image processing in SLAM. In this research we plan to use the Bottlenose camera system from Labforge Inc which has these features of semantic segmentation, deep learning-based hardware accelerated object detection to achieve improved SLAM performance. The additional processing power from Bottlenose
camera processing will help to offload some work from the central processor and result in better real time performance. The research will take existing SLAM solutions and analyze how the Bottlenose camera system can be integrated with them to provide better and improved SLAM solutions

Faculty Supervisor:

Mohamed Atia

Student:

Partner:

Labforge Inc.

Discipline:

Engineering

Sector:

Manufacturing

University:

Carleton University

Program: