Transportation assets are an important part of the road infrastructure and are needed to ensure smooth and safe traffic flow. A well-managed inventory of all assets on a road network is beneficial to agencies responsible for maintaining and managing the transportation assets by making data-driven asset management decisions. A comprehensive inventory of transportation assets will reduce the potential for liability associated with outdated, inappropriately placed, or missing assets.
Most transportation agencies (or the service providers under contract) conduct a semi-automated process, in which the images are automatically collected by the survey vehicle and manually processed in the office. The multi-purpose survey vehicles typically collect data and images for at least one week before they either go back to the office or ship their hard drives to the office. The data and images are then uploaded to a central network, or more recently online cloud, for sharing across the organization. Due to the large file size of high-resolution images, the data upload typically takes several days depending on the connection speed. After data upload, trained technicians use point-and-click software to manually detect and classify the transportation assets.
Our solution integrates the state-of-the-art AI technologies within a practice-ready framework for automated real-time identification of transportation assets from roadway images as listed in the Model Inventory of Roadway Elements (MIRE) which is recommended by FHWA. Apart from MIRE, the system can be modified to accommodate the sign conventions of any other region.
In the modernized process, the data and image collection are conducted according to the existing practice of the agency (or their service provider). However, a plug-and-play edge device with the trained deep learning (DL) algorithm will be connected to the vehicle’s host computer. The edge device connects to the host computer via USB to provide high performance DL inferencing, freeing the host computing power for other tasks such as routing and image acquisition.
Through this proposed framework, the data is processed onboard the survey vehicle. When the hard drives arrive at the office, the processed data (as opposed to raw data) is uploaded to the cloud. This way, the data processing technicians can conduct quality control (QC) of a subset of the processed data, which is a much better use of their costly time than manually pointing to every asset and selecting the correct classification. The results of this QC process can be used to further train the ML algorithm and improve its effectiveness and reliability as more data is collected.
Once the QC process has concluded, the asset inventory data is delivered to the asset management department within the highway agency. Typically, a quality acceptance routine is recommended to accept or reject the delivered data based on testing a subset of the data. Following quality acceptance, the asset inventory data is visualized to facilitate data-driven decision-making.
Our technology deploys one of the very latest machine learning algorithms on a cutting-edge plug-and-play device for superior effectiveness, efficiency, and reliability. The system uses a plug-and-play edge device (Google Coral USB Accelerator) packaged with a deep learning neural network to quickly and accurately process data as soon as it is collected.
iENGINEERING AI technology can be enhanced to provide information such as lane detection, driving area segmentation, pavement marking classification, guardrail detection, and other infrastructure assets inventory data. It also supports edge devices that can be used for plug-and-play integration with mobile devices, cameras, drones, and other offline data collection systems to bring real-time AI capabilities such as autonomous navigation, 3D tracking, scene understanding, and asset detection to such interfaces. The same technology is also being further developed to automatically collect highway assets inventory using LiDAR data, which will provide more accurate measurements of asset dimensions and location