Using AI for automated analysis of road inspection videos

Published by Strada Imaging on

Pavement surface inspection is seriously affected by high per-kilometre costs. Equally, Right-of-way (ROW) videos are rarely analysed due to high labour costs.

DeepRoad, Strada Imaging’s software-as-a-service (SaaS) offering, decouples video capture from analysis, allowing the customers the flexibility of performing inspection with their own camera systems and personnel whenever they want.

The videos, and the associated GPS logs, can then be analysed with DeepRoad’s AI, all from the office through a web interface. The customers can selectively analyse data within DeepRoad.

Hence, they are in complete control of what they want to analyse and when. In marked contrast to the other automated analysis solutions for pavement surface analysis on the market, DeepRoad interfaces are set up to provide the customer with intermediate outcomes of its analysis and conclusions. This level of transparency is required to build confidence in the automation.

One module of DeepRoad automatically detects and characterises/classifies pavement distresses such as cracks, potholes, ravelling, and patching. A second module within DeepRoad detects and geo-locates linear transportation assets like lampposts, traffic signs, marker posts, etc. Depending on their needs, the end users can pay and access a specific analysis service module.

A serious deficiency in most visual road inspection software offerings in the market, except TRACS and SCANNER, is that they cannot measure distress dimensions. For instance, key parameters such as rutting depth and cracking area that determine the condition of the pavement, can only be measured with laser scanners.

Laser scanners require high capital expenditure and therefore are continually being replaced by other cheaper technologies. DeepRoad’s industry-leading AI solution entirely removes the need for laser scanners, measuring critical distress parameters directly from video footage from either one or multiple cameras. Linear assets are inventorized and mapped in 3D with their GPS localization.

DeepRoad can also automatically recognise more than 580 traffic sign types used in the UK. The biggest challenge that Strada Imaging tackled in developing DeepRoad is training a robust AI that can analyse videos obtained from different camera systems. The solution is the result of years of multi-disciplinary R&D, within Strada Imaging and its academic collaborators, spanning 3D imaging, AI, and software engineering.

The pavement surface module of DeepRoad boasts of the following (subject to a high-quality video input):

1. Automatic identification of potholes, cracks, ravelling, patching, joints, ponding, etc

2. Measurement of pothole profiles to a 3mm accuracy in 3D (perimeter, area, depth, etc)

3. Measurement of rut depth to a 3mm accuracy

4. Cracking length and density measured to an accuracy of 10 mm

5. Classifies ravelling into 3 categories

6. Overlaying/superimposing of defect map on local road surface, locating defects with 2 mm accuracy

The linear assets service module of DeepRoad:

1. Identifies lamp posts, traffic signs, marker posts, bus stops, and gully covers

2. Identifies 580 UK traffic sign types individually

3. Locates assets to a resolution better than that of GPS

4. Maps parked cars, enabling resurveys in those areas

5. Traffic cone identification and GPS-style localization

DeepRoad has a 24-hour all-year-around availability. The software makes use of the Google Cloud Platform ensuring 100% uptime.

Categories: News