Automatic Pothole Detection by Different Multispectral Band Combinations

The road is one of the main infrastructures that play a significant role in supporting the economy and human social development. Therefore, this study was aimed to automate potholes extraction using UAV multispectral images. The objective of this study was to identify the band combination for pothole extractions based on multispectral sensors and analyse the 2D pothole model with the actual measurement. The methodology was divided into four stages- planning, data collection, data processing, and data analysis. The DJI Phantom 4 aerial vehicle with a Sequoia+ Multispectral Camera was used to perform the data acquisition. All data acquisition was operated by three different software, namely Pix4Dmapper software, ArcGIS software, and SAGA GIS software. The final outputs generated from this study were orthophoto, combination bands, and the pothole extraction area. Thirteen combination bands, including a single band, were made to identify the best combination band for pothole extraction. There are two types of combination bands- the two-layer combination band and the three-layer combination band. Two types of potholes were examined, namely Pothole A and Pothole B; for Pothole A, out of the 13 combination bands, the best combination band for pothole extraction was the combination of green, red, and red-edge bands, with a zero-difference value between the actual measurement and computed area. Meanwhile, for Pothole B, the best combination band was the NIR, red, and red-edge bands, with a 0.075 m(2) difference value between the actual measurement and computed area.