This project aims to develop the deep convolutional neural networks for supporting the perception models to detect rails paths, obstacles, wayside elements and surrounding environments using camera and lidar scanning data. The developed perception modules will assist in autonomous train navigation and control.
This $2.6M project funded by Teledyne Optech and the NSERC aims to develop deep learning-based computer vision systems for improving the efficiency and robustness of post-data acquisition process such as object detection, segmentation, environment modelling and utility network modeling with Teledyne Optech’s cutting-edge mobile mapping systems, and a novel SLAM (Simultaneous Localization and Mapping) system.
In this project, we study the role of high-level geometric topological cues such as 3D primitive representation of indoor corridors to improve the performance of simultaneous localization and modelling (SLAM) using a low-cost monocular camera. We aim to develop computer vision systems for automatically reconstructing 3D manhattan models to represent an indoor corridor from monocular imagery, and integrate it within SLAM framework.
Q-Drone is a research-oriented unmanned aerial vehicle (UAV), which provides objective path planning to meet the requirements of critical infrastructure mapping such as the complete and homogeneous quality of structure-from-motion results and object waypoint generation for securing a novel view of anomalies. Q-Drone will provide a novel data-driven capability to use ultra-wideband sensors in cooperation with visual sensors (IR and RGB camera) with augmented positioning using deep convolutional neural networks. We provide a unique benchmark for the research communities, called “Q-Drone UWB benchmark”.
Grant Agency: NSERC CREATE – Data Analytics and Visualization; ORF Infrastructure – Advanced Disaster, Emergency, Rapid Response Simulation (ADERSIM)
3D City Modelling
Dr. Jaewoo Jung published his research work on 3D building modelling methods in the ISPRS Journal (doi.org/10.1016/j.isprsjprs.2019.01.003). This journal suggests an innovative method for automatically updating an existing 3D rooftop model by aligning temporal imagery. This progressive modelling pipeline is developed based on Markov Chain Monte Carlo coupled with simulated annealing. The performance evaluation conducted with the ISPRS benchmark data proves the potential of our progressive modelling approach as an effective tool for updating large-scale city models.