Roman Giroux
Software Engineering
Edward DiBella
Software Engineering
Dr. Siddhartha Bhattacharyya
Department of EECS
Program Chair | CSSE
Michael Nicol
Computer Science
Our Goal
Our goal is to expand the scalability of the AssistTaxi annotation pipeline that stretches beyond the current limitations to include broader weather, obstructions, and time of day scenarios. The original CDLEM pipeline tool demands 120 contour definitions to label only sixty thousand frames and included miss labels of centerlines, and incomplete detections requiring constant human intervention.
The disconnect between speed and reliability is our main motivation in improving AssistTaxi. The use of ALINA cut the processing time almost in half by applying reusable regions of interest; however, the same runway failures occurred.
We aim to create a reliable, fast, and safe way of labeling. This removes the bottleneck for downstream researchers, annotators, and engineers who will use this dataset.
First Semester Milestones
Milestone 1 (Sep 28):
Create a high-level requirements document that outlines an improved data generation, labeling, and model training
Research additional model architectures and improvements upon the existing models. This literature review would look for ways to improve the accuracy of the system
Create a high-level architecture diagram, documenting both the current system and an improved proposed architecture to implement in following milestones
Milestone 2 (Oct 26):
Create an initial Minimal Viable Product (MVP) demonstration of the improved architecture or model on a select dataset
Test MVP architecture and create a harness to compare and contrast the performance of the new system against the legacy system.
Receive and implement feedback from stakeholders in the project.
Investigate X-Plane for synthetic dataset generation for auto-labeling.
Create Milestone 2 presentation for senior design.
Milestone 3 (Nov 23):
Transition MVP architecture into an end product with well defined low-level requirements. Document low-level architecture.
Integrate synthetic dataset generation into pipeline if determined to be viable for usage.
Receive and implement feedback from stakeholders in the project.
Polish semester one architecture and define requirements for the next semester.
Create Milestone 3 presentation for senior design.