A photogrammetry-based image registration method for multi-camera systems – With applications in images of a tree crop. (October 2018)

• Record Type: Journal Article
• Authors: Gan, Hao; Lee, Won Suk; Alchanatis, Victor
• Is Part Of: Biosystems engineering. Volume 174(2018)
• Page Start: 89
• Page End: 106
• Publication Date: 2018-10
• Subjects: Photogrammetry -- Precision agriculture -- Registration -- Remote sensing -- Thermal; Bioengineering -- Periodicals; Agricultural engineering -- Periodicals; Biological systems -- Periodicals; Génie rural -- Périodiques; Systèmes biologiques -- Périodiques; 631
• DOI: 10.1016/j.biosystemseng.2018.06.013
• Languages: English
• ISSNs: 1537-5110

Abstract

Abstract : In precision agriculture, estimating crop yield using remote sensing techniques is an active research field. To achieve high accuracies, researchers frequently combined different imaging sources, such as colour (Red, Green, Blue [RGB]) images, thermal images, and near-infrared images. However, fusing information from those images has been a difficult task. Therefore, accurate image registration methods are necessary. This study aimed to develop a thermal-colour camera system which will register thermal images with colour images of tree canopies in preparation of information fusion and fruit detection. The registration method created in this study was based on photogrammetry. In preparation of registration, a camera system was built, consisting of a thermal camera and two colour cameras. Camera calibration, image intersection, and space resection were combined in a single step named 'stereo-calibration', to compute cameras' parameters and poses. Speeded-up robust features (SURF) were used to find points of interest from colour images. Random sample consensus (RANSAC) was utilised to search for optimal homography transforms between thermal and colour images. In addition, this study created a procedure for accurate registrations of regions of interest in thermal-colour image pairs, utilising structural similarity (SSIM) index. The proposed method offered pixel-level registration accuracy and achieved an average accuracy of 3 pixels in 640 × 480 – pixel citrus canopy images. Highlights: Complex scenes were segmented into simpler ones for matching regions of interest. Photogrammetric and feature matching methods was utilised for accurate registrations. The method could register multimodal imaging systems of different types of cameras.