For an automated street marking detection system, a diverse dataset is essential. For collecting the images, either drones or special road vehicles equipped with necessary hardware for data collection may be used. Drones capture spatial data, but issues like shadows from trees and buildings can affect the quality of image dataset. Similarly, sensors data collected from the road vehicles face challenges like obstruction by moving vehicles and low information density in markings. Traditional image-processing methods rely on individual features, but they struggle with distortions, lighting changes, and distractions. A deep-learning-based AI (Artificial Intelligence) model has the possibility to offer a solution by learning robust data representations.
For AI-based street marking detection, combining the data collected from drones and road vehicles has the potential to provide a more accurate assessment. Drones offer spatial data via ArcGIS, while public-sector/municipal vehicles, such as garbage trucks and police cars (when equipped with necessary sensors), will provide terrestrial data. This dual approach will ensure a comprehensive coverage from different angles and heights, allowing for precise monitoring of street markings once the model is trained. Defined routes of municipal garbage trucks enable consistent tracking of street markings, while police vehicles’ variable paths (usually moving as per the message received) provide additional dataset and can even be used for verifying the condition of street markings in an area if needed. The terrestrial information obtained from the chosen police cars and garbage trucks (operating almost regularly) will assist in detecting the distortions in the street markings quicker as compared to spatial data which is usually collected by drones twice or thrice a year.
Real-time data transmission via 5G technology (leveraging its high-speed data transmission, low latency, and enhanced connectivity), using custom-designed cavity filters will ensure efficient communication between vehicles-to-vehicles (V2V) and the vehicle-to-infrastructure (V2I) or to control center. Having such an integrated system will reduce the need for visual inspections, streamline maintenance and optimizing decision-making through AI and ML models trained on both spatial and terrestrial data (separate for both kinds of data).
The proposed solution will comprise of designing customized 5G cavity (band pass) filters and tuning them by fully automated filter tuning robot developed by the main contact of this solution1. These cavity filters need to be tuned to compensate the design tolerances and variation in material properties once they are assembled. A perfectly tuned filter would allow only the desired frequency band to pass through them. In our solution, the communication channel will use the sub-6 GHz band (FR1 band). Commercial telecom companies operate within the 3.3 GHz to 3.9 GHz range (N77, N78, N79 bands), while the latest Wi-Fi band operates between 5.7 GHz to 5.9 GHz. Our vehicles will operate beyond these commercial bands. We will design 5G filters to operate at 30 dBm (approx. 1 Watt) power, effectively blocking high-power signals (45 dBm or 40 Watt) and discarding spurious harmonic signals. Repeaters will be strategically placed throughout the streets to ensure the required signal quality reaches the control center. This approach minimizes manual intervention, enhances efficiency, and improves the reliability of street marking maintenance.
