Rapid developments in the research of artificial intelligence (AI) make it possible to transfer tasks that are repetitive and tedious but at the same time complex in nature and attention- and precision-demanding from humans to machines. Moreover, using AI makes procedures of the given tasks quicker and therefore cost-effective. Road inspection for deterioration can be one of these tedious yet complex tasks. Could a machine armed with an AI and state-of-the-art hardware save road officers from this assignment?
Doing it the good old-fashioned way is costly
Pavement tends to deteriorate with time under the influence of traffic and pressure from the environment, such as varying weather conditions and ground movements. Efficient and timely road inspection is one of the key elements of a successful pavement management system. However, periodical human visual inspection of the road surface tends to be costly and time-consuming. This is because, first, the inspection has to be done quite slowly so that the human eye could detect possible defects. Second, two persons are needed as humans are able to concentrate on only one task at a time. Driving a car is a highly complex and attention-demanding task and it’s not safe, if not even impossible, to inspect the road simultaneously with driving a car. Third, all the gathered data is entered manually in the system and only then it is possible to analyze it. However, with the help of an AI, these costs can be reduced substantially.
How does AI work in defect detection?
The use of AI in defect detection is more and more widespread among companies that offer road inspection software. Well, how does it work?
As with any AI-powered software, the AI is first trained on thousands of images to be able to distinguish defected pavement, such as potholes, edge defects, simple cracks, network cracking, and weathering, from the acceptable road surfaces. For training, two types of input data are used: new input images and annotated input images. The AI learns from the manually annotated data so that it is able to distinguish different types of road defects on new data. When the training is finished, the AI-powered software is tested for success rate by comparing its detection results with the results of an experienced human observer. When the success rate is high enough, then the software is ready to be used. If it’s not, more training will take place to reach a higher accuracy level. When the training process has ended and the software is ready to be used, all that needs to be done is to gather the input data that has to be analyzed and let the AI processing do the rest of the work.
One of several approaches to gathering input data for defect detection is using smartphones. With this method, and depending on the company, road officers just have to download an application, position a smartphone on the windshield, and the defect detection process can be started. Although user-friendly and simple to use, there are downsides to this kind of approach.
The downsides of the smartphone-based approach
Due to their relatively small size, smartphones also have small image sensors. Small sensors mean poor resolution and noise in low-light or other difficult shooting conditions. In addition, their geo-data is quite inaccurate as they generally use AGPS to locate their position. This, for example, can cause a horizontal position error of 7–13 m in an urban environment (tested on iPhone 6). Thus, while smartphones can be sufficient for simple data collection, they lack in location accuracy and image quality. This, however, can be a problem because for the best end results, the input data has to be of the highest quality.
How does EyeVi do it?
EyeVi Technologies Ltd. has developed an AI pavement defect detection system that consists of several submodules. The input data is gathered using state-of-the-art technical apparatus that is portable and mountable to any type of car. The apparatus includes a 360° panoramic camera, LIDAR scanner, GNSS/INS system, and data collection computer with software. This allows to easily gather various types of data at a highway speed.
The EyeVi pavement distress AI has been trained in collaboration with the Estonian Road Authority. Through the processing of 360° camera output data, we generated orthomosaics that the Estonian Road Authority used to manually digitize 21,000 km worth of road defects. This input data was then used to train the AI and now it can surpass the capabilities of manual digitizers and achieve constant and uniform quality level throughout the road network. Since the whole process is automatic, it speeds up the data gathering and analyzes process and can heavily reduce the costs per kilometer. But this is not all that can be offered. In addition, the EyeVi software platform generates panoramic images and 3D pointcloud that can be used to detect various aspects of road furniture, such as traffic signs, road barriers, fences, and profiles. This way EyeVi can digitize everything necessary and provides an excellent quality level for a wide range of use cases.
So, with the right equipment, the possibilities of AI use are even wider than just helping out the road officers in their defect detection task by charting the cracks and other signs of deterioration in the pavement.