Automatic Traffic Counting with Computer Vision

Preliminary Tests Without Active Recording

In order to generate quality data to be used in research, the aCar team has constructed traffic counters. The aim is to set up the counters in rural areas of Sub-Saharan Africa, as mobility data is particularly sparse in those regions.

The Traffic Counters

A counter is a self-sufficient supply capsule that can generate its own power by means of a solar panel. The actual measurements are made by a smartphone running object recognition software. An external battery enables the counter to continue to function even in the event of unfavorable weather conditions and decreased solar output.

The Object Recognition Software

This image shows the object recognition software in action at a roundabout in Sub-Saharan Africa. The software has to be trained with images to recognize specific targets via machine learning. It is of great importance to use images in the correct setting in order to achieve a high rate of detection.

The Testing Process

In order to test the feasibility of the concept, a prototype device has been mounted on the property of the Chair of Automotive Technology at the Technical University of Munich. The testing and validation is to be completed in 3 phases:

Phase 1 (in progress):

Testing the energy output of the solar panel. In this phase, no smartphone or camera is attached to the device. The solar charge controller, solar panel and power electronics are active. The goal is to complete a full charge of the battery.

Phase 2:

Testing the power system under load. In this phase, a smartphone is set up and connected to the power of the external battery. Note: The camera is pointed inward so that recording the surrounding area is impossible. This means that the smarphone serves as a power drain and a monitoring device. The goal is to record the SOC (state of charge) of the smartphone and assess the overall availibility and status of all systems.

Phase 3:

Automatic Traffic counting: The smartphone is pointed outward and the camera is activated. The goal of this phase is to test the complete system including the object recognition software (e.g. 2 pedestrian, 3 bicyclists, 1 car)

Future Outlook:

After successfully validating the counter, a small batch is going to be constructed and shipped off to Yamoussoukro in the Ivory Coast. Here, they are to be mounted in key locations of interest and begin data collection for the aCar Mobility Project.

Leave a Reply

Your email address will not be published. Required fields are marked *