Semi-automated roadside image data collection

Pilger, Neal; Berg, Aaron; Joosse, Pamela · 2019 · Crossref

DOI: 10.5194/gi-2019-20

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the logistical and financial challenges associated with collecting ground-truth data for remote sensing of agricultural land use, specifically post-harvest tillage and cover cropping practices. Traditional methods rely on labor-intensive, in-field physical measurements that are costly and limited in spatial coverage. Furthermore, optical satellite remote sensing is often hindered by cloud cover during the critical non-growing seasons (late autumn and early spring) when these practices are assessed. The authors developed and tested a semi-automated mobile roadside image collection system to provide a low-cost, efficient alternative for generating geo-referenced validation data that is insensitive to atmospheric conditions. The methodology involved deploying a vehicle-mounted imaging system in Elgin and Essex counties in southwestern Ontario. The system utilized Garmin VIRB XE cameras and a modified GoPro HERO camera with near-infrared capabilities, mounted on extension poles to capture oblique views of adjacent fields. Data were collected in May 2016 at driving speeds of 40–45 km/h, with shutter actuation calibrated to ensure multiple images per field. A total of 18,462 images were acquired, sorted, and geometrically rectified to remove lens distortion. These images were geo-referenced using embedded GPS data and integrated into a GIS environment. The roadside imagery was validated against detailed in-field ground verification data from 114 research sites. The ground truth data were derived from nadir-view photos analyzed using a 10x10 grid sampling technique to quantify residue cover, categorizing fields into conventional tillage (<30% residue), conservation tillage (30–60%), no-till (>60%), or green cover. The results demonstrated a high level of correspondence between the mobile roadside imagery and the in-field ground truth data, with an overall accuracy of 93%. The confusion matrix revealed that misclassifications primarily occurred between conservation tillage and no-till categories. The authors attribute this to the visual similarity of these classes from a roadside perspective and the geometric differences between oblique and nadir views, which can exaggerate residue visibility. However, since both conservation and no-till practices meet the threshold for soil erosion protection (>30% cover), these discrepancies are considered acceptable for policy and program decision-making. The system proved capable of surveying up to 500 fields per hour, significantly outperforming traditional methods that cover only 5–10 fields in the same timeframe. The study concludes that semi-automated roadside image collection is a robust, cost-effective method for monitoring agricultural soil cover. It offers significant advantages over satellite remote sensing by operating independently of cloud cover and over traditional field surveys by reducing personnel costs and increasing spatial coverage. This approach facilitates routine, large-scale monitoring of tillage practices and can support time-series analysis for assessing climate adaptations and regulatory impacts. The authors suggest this method is suitable for adoption by government and industry organizations for more efficient and reliable assessment of agricultural land management practices.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement