Distortion and mosaicking of close-up multi-spectral images

Summary

In precision agriculture (PA), vegetation indices (VI) are commonly used to evaluate the health of crops with the use of multi-spectral cameras. These are often mounted on drones and used at high altitudes, where the translation of focal points and subsequent changes in perspective do not pose any difficulties. In proximal sensing, however, these translations and distortions pose a significant challenge on data processing. In this work, a Parrot Sequoia camera was mounted at a fixed height of 1.2 m and used at 3 sec and 1.5 sec intervals on an organic strip-cropping field beneath. Reference imaging revealed that the multi-spectral lenses suffer from significant barrel distortion of 30%, while the higher resolution RGB lens has a barely distinguishable 1% pincushion distortion. The subsequent field images were stitched together using an open-source panorama software to automatically detect and correct distortions. The resulting mosaics were then shifted to correct the relative position of the separate lenses, allowing for VI calculation with mm accuracy. This method allows for analysing single plants for inter-crop and even intra-crop variation, allowing the automation of tasks in an organic multicrop environment.