Each state is mandated by federal transportation authorities to use traffic management systems (TMS) to track the number of cars traveling on the road, the type of vehicle and its weight. That data collectively helps promote efficient travel and keep people safer.
However, the systems in place now are expensive, and depending on the type installed, not always possible to move.
The current options are intrusive sensors embedded into the pavement, nonintrusive sensors that sit on top of or over the roadways and off-roadway sensors, such as satellites. However, a research team at the University of Memphis proposes a different system using Wi-Fi signals and deep learning. They say the system they’ve developed is low-cost and portable, potentially making it an excellent TMS to use in the future.
How Does It Work?
The technology is a nonintrusive system called DeepWiTraffic that tracks and classifies vehicles based on the channel state information (CSI) of the automobiles that move past. CSI encompasses the properties of a wireless communication link.
The team figured out a method of determining how the vehicles that go past cause changes in specific CSI information, including spatial and time correlations. That data is used to classify the kinds of vehicles and how much they weigh.
A deep learning algorithm makes up the second component of this system. First, the researchers made a convolutional neural network (CNN).
It’s a kind most often used for visual image processing, and the one the researchers created captures optimal data from the CSI information. Next, they trained a vehicle classification model using preprocessed CSI data.
When tweaking that vehicle classification model, they also aimed to reduce the effects of things that could lower accuracy, such as the presence of people or nearby vehicles. Ultimately, the classification capability distinguished among five kinds of vehicles: motorcycles, SUVs, passenger cars, trucks and large trucks.
An Impressively Accurate Option
During the testing of this new system, the researchers used ground truth video data to ascertain how well the setup worked. They also ran a testing period for about 120 hours over a month. In total, the vehicle detection aspect was 99.4 percent accurate, and the component that classified vehicles was correct 91.1 percent of the time.
However, they clarified that it was difficult for the deep learning algorithm to classify vehicles that looked similar, such as pickup trucks and SUVs. In those cases, the classification accuracy measurement dropped to just over 83 percent. Despite that minor pitfall, the team believes their system offers substantial potential.
Machine Learning and Deep Learning Innovations Show Promise
Perhaps one of the most impressive factors of this new TMS is how inexpensive it is. The University of Memphis team says it costs about $1,000 to deploy. In contrast, other systems used now may represent tens of thousands of dollars in expenses, depending on the type used and the amount of roadway covered.
Since machine learning and deep learning algorithms help achieve feats like this one, college students are increasingly eager to pursue computer science master’s degrees and then put their efforts toward niches like machine learning. Their enthusiasm could be instrumental in pioneering future developments.
It’s also a positive sign that machine learning is among the most in-demand fields, as evidenced by the rising number of job postings for machine learning opportunities and the growing market worth of the overall industry. People who want careers in the field will have opportunities to work on projects like the TMS above while enjoying the likelihood of excellent job security.
How Could This TMS Improve Traffic Flow?
According to the full research paper about this new TMS, there is an online mode that works for real-time classification.
Besides offering historical data such as how many large trucks pass through an area between midnight and 4 a.m. per day on average, the up-to-the-minute classification option could theoretically manage higher levels of traffic coming into an area, like for a special event.
Traffic managers might also use the technology to gather data that suggests the need to change traffic routes for safety reasons, either on a temporary or permanent basis.
An Especially Practical Solution for Rural Highways
A concluding section of the research about this innovation highlighted how its low cost makes it a particularly good fit for the sprawling miles of rural roadways.
In those cases, it can be exceptionally costly to implement and maintain TMS. For now, people can only wait and see how big of an impact this improvement might have, but it seems to be a significant one. DeepWiTraffic has the potential to greatly improve transportation safety and efficiency.
