Technology description

Detection and segmentation refinement

We use modular architecture to easily add new types of distresses and pavements. Our current detector may detect and separate even interconnected different types of cracks from each other. It is important for correct PCI estimation due to varied physical genesis and different effects on pavement conditions. Potholes are detected separately.

Measurement

Our application measure distresses only from ego-vehicle (a vehicle which is used for video-registration of a road) based camera without the usage of additional equipment. Measurements are used for severity estimation, classification, and overall statistics calculation. Roadly created the SLAM engine which allows us to model a 3D environment around the ego-vehicle. It is used for direct estimation of the length and area of distresses.

Some measurements may have too strict requirements for accuracy, e.g. crack width/depth estimation (starting from 6mm) that is infeasible by camera-based visual registration. We have replaced such steps with integral AI-based approaches that estimate the whole visual context (e.g. cracking level, connectivity, spalling, etc) that highly correlate with width.

Sub-type and severity classification

We use an ensemble of AI-based models for refined classification of distress type and severity estimation. All the models may be significantly improved and adapted by online data collection from real customers. Also, crack topology modeling is used for more accurate estimation.

Road modeling

PCI is significantly impacted by proper road modeling. We use SLAM for the initial 3D modeling of the road. It is combined with video-based visual segmentation of the road and GIS data (maps) to accurately model the road and correctly split it into sample units and sections. A subdivision of a pavement section has a standard size range: 225 contiguous square meters +- 90 m^2, if the pavement is not evenly divided by 225 or to accommodate specific field conditions.

PCI calculation

We use a standard estimation procedure for the calculation of the final PCI. The principal distinction of our approach from manual PCI estimation is that we do not use sub-sampling of units to estimate the whole section statistics. According to our approach which is automatic by its nature, we could process every sample unit of a section. Also, we could vary the road sectioning strategy to get more granular or bigger sample units by request from our customers.

Measurement procedure for ACP:

- Inspect a sample unit

- Record branch and section number

- Sketch the unit, including orientation

- Record a sample unit size

- Distress inspection

- Define Severity

- Measure

- Calculate Quantity

- Calculate PCI

We modeled the statistics aggregation procedure following steps:

1. Group all distresses by type (below in the table), severity (H/M/L), and measurement units (area (sq meters), linear (meters), number of occurrences).

Figure 1. Example: Survey data sheet (Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys, p.29)

2. Get the percent density of each distress type and severity by dividing the distresses’ corresponding measurements sum by:

- Total area of the sample unit

- Or the total length of the sample unit

3. Calculate Deduct Value (DV) for each type of distress and severity level combination

Figure 2. Example: DV calculation for an alligator cracking (Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys, p.29)

4. Determine the maximum corrected deduct value (CDV) from individual DVs

Figure 3. Example: maximum CDV calculation (Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys, p.48)

5. Calculate PCI

Figure 4. PCI formulas

Conclusion

To conclude, we developed a modular architecture of application for pavement condition estimation which follows the ASTM standard. We have covered the most common pavement types and distress types and we plan to add the rest in the nearest future. The following improvements of AI models and the addition of new categories will be fully compliable with previous versions of PCI estimations and current expert-based evaluations of road sections.