Improving the Airline Boarding Process Through Deep Learning

Most frequent travelers lose hundreds of hours due to take-off delays. One of the biggest factors in those delays is the amount of hand luggage that passengers can bring into the cabin. Therefore, knowing how much luggage must be put on hold before boarding is of the utmost importance both for the airlines and the passengers to ensure a better travel experience.

The SDG Group Advanced Analytics team decided to find a solution to this problem by developing a suitcase detection software based on the YOLOv2 architecture.

We will go through most of the important architectural decisions regarding the algorithm. For instance, adding state-of-the-art components like inception or residual blocks might add more accuracy but prevents the algorithm from running in real time; as with most machine learning tasks, there is always a tradeoff between accuracy and prediction speed.

Most of the current building blocks for object detection will be explained (such as hybrid methods like RCNN, and state-of-the-art solutions like Pyramidal Neural Networks). We will explain how to deploy this solution using an edge device like AWS DeepLens, and how to integrate it with other components of the AWS stack.