ASEE NCS Conference 2019

Full Program »
PDF File
View File

Predictive Queue Times

The project’s aim is to build a monitoring application for human queues. The goal is to detect and identify queues that lack well defined structure, regularity, or organization (e.g. queues that exhibit clusters of people and movement), or queues whose shape is not completely predictable (e.g. queues that are formed spontaneously at different locations and cannot be monitored using fixed motion sensors). A good example of this would be at train stations in front of the ticket booth.

The application will be able to calculate the average wait-time for people in a queue using a simple camera and a neural network to recognize people shapes. The application would abstract these shapes to detect the presence of queues, people clusters, and the movement of people through these queues. It will distinguish people waiting in a queue from passersby. The calculated average wait time can be used in two scenarios. One, it can give near instantaneous prediction of wait time for people before they join a queue. For example, a passenger could check the wait time to buy a ticket before ever leaving their house to help plan when to leave to buy a ticket, removing the fear of missing the train. Two, given the average wait-time over long periods, businesses can make intelligent decisions about staffing and logistics that will improve their services to customers.

This application will have to overcome hurdles such as privacy issues related to the storing and processing of videos containing people’s movements as well as being simple to install and maintain. In the end, the application of this system should benefit the businesses who purchase it, as well as benefit the customers of those businesses.

The proposed application will function based on the publicly available neural network implementation TensorFlow. Specifically, the network will be pre-trained to identify humans from visual input, which will be provided by the attached camera. This human recognition data will then be analyzed by the software to identify and track queues. The queues will be tracked not only based on individuals, but also based on clusters. At various establishments, families and groups of friends will get in line together, but they function as one group for service. Therefore, clusters will be identified and tracked based on their proximity. This will allow the queues to be tracked without gaining inaccurate impressions due to large clusters of people.

Corey McLaughlin
Ohio Northern University
United States

Jacob Hartman
Ohio Northern University
United States


Powered by OpenConf®
Copyright©2002-2018 Zakon Group LLC