Automated Optical Crack Detection in Pavement

Automated Optical Crack Detection in Pavement

At Event 38, we build drones for data collection, sensors and post-processing software. That puts us in the unique position of being able to support the full workflow from flight to results but also gives us all the tools we need to explore novel applications for drone data. Today, we’ll investigate the capability of drones to automatically capture, quantify and report on the presence of cracks in pavement. We’ve written an algorithm that automatically filters cracks from imagery and then calculates their length, width and density.

 

Method

Our algorithm uses ortho-rectified and stitched color imagery only. A series of morphological transformations is made to highlight then quantify the position and shape of each crack while ignoring irrelevant changes in color and texture. Although it would seem helpful, point cloud (elevation) data from LiDAR or photogrammetry would require excessive data collection to resolve height differences at the scale of most common cracks.

 

Capabilities

Using automated image processing only, we are able to separate real cracks in pavement from wet spots, stains, painted lines and tire marks.

DiscolorationWhiteLinesTireTracksCracksDetected

The algorithm can be adjusted to discern cracks of different sizes. Detecting smaller cracks increases the rate of false positive detection, but most can be filtered so that small cracks are only detected when they occur together in groups.

FilterSize

We’ve also created a set of analytics tools that organize data about the cracks into human-readable statistics. These values can be used to compare the density of cracks between areas or over time.

SmallDetect
Estimated Average Width 1.01cm
Length 10.53m
Total Area 14.46m²
Density (linear distance/unit area) 0.7267m-1

 

Limitations

This algorithm works best on light colored, clean pavement. Asphalt and tar hide the contrast of cracks and return few results for cracks less than 4cm wide. Surfaces with deteriorating coatings can also cause too many false positive returns because the peeled layers create edges and shadows that appear as cracks to the algorithm.

Crack2-2

 

Applications

Despite the limitations, there are useful applications for crack detection and monitoring. Because the cracks can be quantified and mapped by density, the data can be used to determine when repairs are due and to monitor deterioration over time. Because the cost of collection and processing is small, it can be applied to many common commercial and industrial applications such as parking lot and private road maintenance.

Data for this study was collected at 35m altitude with a 12MP sensor. Assuming a small, square or rectangular project area and using an Event 38 multirotor mapping drone, data could be collected at a rate of about 1.2 acres per minute of flight. Including battery changes, an operator could cover up to about 45 acres per hour. If the project area is a long, thin strip of pavement such as bridges, roads, highways or runways, a fixed wing E384 or E386 could be used to cover significant distances quickly. An E384 could cover up to 21km linear distance over the course of an 80 minute flight making three passes at different angles.

We are encouraged by the promising early results from this algorithm but more user feedback is needed. If you can collect high resolution data or are interested in a joint research project, please contact us to discuss your application.


Want to stay up to date on our latest findings? Sign up for our newsletter!