Vision based stockpile inventory measurement using uncrewed aerial systems

Abstract

Monitoring a stockpile plays a vital role in material inventories at the State Departments of Transportation (DOTs). Various technologies, such as Total Stations (TST), Light Detection and Ranging (LiDAR), and Global Positioning Systems, are conventionally used to obtain stockpile volumes; however, DOTs seek a faster, safer way to obtain an object’s volume with minimal workforce training. Uncrewed Aircraft Systems (UAS), coupled with visual imagery, have the potential to address these limitations; however, UAS visual has not been effectively developed to account for flight parameters in measurements, such as Ground Sampling Distance (GSD). Images of regular and irregular objects were collected in several flights to measure their geometries. The measurements were performed using a computer vision algorithm and a common commercially available photogrammetry tool (Pix4D) as UAS visual and UAS LiDAR. The results indicated that UAS visual is a viable technology that provides consistently accurate measurements of stockpiles of various sizes. The authors used Pix4D to measure volumes ranging from 0.45 m³ to 2838 m³ with errors ranging from 4 % to 6 %. The results indicated that ensuring a GSD value of 0.80 cm in visual imagery can lead to accurate volumetric measurements of irregular objects. To reduce the processing time, a deep leaning-based point cloud classification model was developed to detect the objects of interest, stockpiles, and separate them from irrelevant objects. The average volume difference between the volume predicted using Pix4D and point cloud classification was less than 5.5 %. Finally, we compared the advantages and challenges of UAS with traditional methods and UAS LiDAR in terms of data collection time, cost, limitations, and safety. The results demonstrate that using UAS for stockpile volume measurement is safer and more time-consuming and cost-effective.