PCI calculation methodology

Roadly’s PCI calculation methodology is based on the ASTM standard. We literally follow the estimation process from the standard, but replace the manual distress evaluation procedures with AI-based models.

PCI is calculated from different statistics of defects and distresses present in a Sample Unit of road. The index is affected by the number of defects detected, their severity level, and corresponding metric measurements (length, width, or area). We followed the “Distress identification manual for the Long-Term Pavement Performance Program” to train our AI to detect different types of distresses and their severity level. Linear measurements are performed by our visual-based algorithm from input frames.

Currently, Roadly confidently identifies five types of cracking (transversal, longitudinal, edge, alligator/fatigue, and block) followed by potholes. These types of distresses have the most significant effect on PCI total rating in the majority of cases. However, we are also developing AI-based models for patch deterioration, rutting, shoving, and other distress detection and identification. It should be noted that we have developed robust detection models only for the pavement with asphalt concrete surfaces type, which is the most common pavement type. Other types (jointed Portland cement concrete surfaces and continuously reinforced concrete surfaces) are under active development.

Feature	Expert-based PCI calculation	Roadly PCI calculation approach
General
Method	Expert-based + special equipment (lidars, geodetic equipment)	Only video from a phone installed in vehicle
PCI calculation methodology	ASTM	ASTM
Reporting	ASTM	PCI + distress statistics per sample unit
Sample units
Road splitting into sample units	225 ± 90 square meters	Arbitrary, defined by the customer, may be 225 square meters
Sample unit area estimation	Based on direct width/length estimation of road	Visual estimated model + GIS data
Sample units sub-sampling (number of sample units estimated from total road area)	Sub-sampling (All sample units in the section may be inspected but this is usually precluded for routine management purposes by available manpower, funds, and time. Total sampling, however, is desirable for project analysis to help estimate maintenance and repair quantities)	All
Distress estimation
Types of defects	Based on survey provider	5 types of cracks + Potholes, the rest under development
Measurements	Width: std error from 1mm Length: std error from 1mm Depth: std error from 1mm Area: based on simplified contours (circles, rectangles)	Length: std error from 3cm Width/Depth: not estimated directly, AI-based approach estimation by visual context Area: std error from 9cm2
Classification	Methodology based expertise	AI-based approach combined with distress topology modeling
Severity	Methodology based expertise	Methodology based models + AI