Edge Computing Based on Convolutional Neural Network for Passenger Counting: A Case Study in Guadalajara, Mexico

Area of Science:

• Computer Science
• Transportation Engineering

Background:

• Public transport in the Guadalajara Metropolitan Area faces challenges with fixed route frequencies unable to meet dynamic passenger demand, leading to long waiting times.
• Addressing this requires intelligent systems capable of adapting to real-time user variations.

Purpose of the Study:

• To design and implement an embedded system module for accurate passenger counting in public transport.
• To improve public transport service quality by enabling data-driven adjustments to route frequencies.
• To provide a foundational solution for intelligent public transportation systems in Mexico.

Main Methods:

• Developed and experimentally validated an embedded passenger counting system using image analysis to determine user numbers.
• Generated two novel datasets specifically for training and testing the CSRNet algorithm on images from Mexican public transport systems.
• Implemented the passenger counting system on a Jetson Nano development board for practical application.

Main Results:

• The designed system successfully counts passengers from images and transmits this data to a server.
• New datasets were created to enhance the performance of the CSRNet algorithm in Mexican urban transit contexts.
• A functional hardware implementation of the passenger counting system was achieved using the Jetson Nano.

Conclusions:

• The developed embedded system provides a viable solution for real-time passenger counting in public transport.
• This technology can significantly contribute to optimizing public transport operations and enhancing user experience.
• The creation of localized datasets and hardware implementation pave the way for smarter urban mobility in Mexico.