A team of researchers at the University of Memphis in the US have built a portable low-cost Traffic Monitoring Systems (TMS) named DeepWiTraffic that utilizes Deep Learning and Wi-Fi to deal with the issue of costs while placing the TMS. The systems are installed across the nation to accrue traffic data that contain insights about the range of vehicles, its density, speed, and vehicle class to advance safety and efficiency.
Currently, the costs of installing a single TMS on a two-lane roadway comprised USD 25,000. So in an effort to low these costs, DeepWiTraffic embraced non-intrusive method utilized sensors and wireless channels to epitomize vehicles. As the system depends on WiFi to spread Channel State Information (CSI), deep learning is utilized to instruct the vehicle classification model on the basis of effectively preprocessed CSI data input. CSI, in wireless communication, is essentially a channel asset that explains how signals are conveyed from a transmitter and to a receiver and vice versa. CSI channels, in DeepWiTraffic, were leveraged to communicate insights about changes caused by passing vehicles and even variations in environment noises for successful vehicle identification. So, to train the system, the researchers conducted wide-ranging real-world tests utilizing CSI data and ground truth videos which they accumulated more than a time span of 120 hours. As a result, they gained outcomes with an accuracy rate of 99.4 percent where the device seemed able to detect passing vehicles into five vehicle types, including passenger cars, motorcycles, SUVs, pickup trucks, and large trucks.
As per the research paper, the trails were done with huge combinations of hyper-parameters in training the CNN model (Convulational Neural Network) to advance the accuracy of detection. The researchers expect to install their solution in the US rural highways that stretched up to 119,247 miles. The researchers noted in a report that DeepWiTraffic will contribute to deploying a wide range of TMS with lower costs.