Automated, high-throughput image calibration for parallel-laser photogrammetry

Abstract

Parallel-laser photogrammetry is growing in popularity as a way to collect non-invasive body size data from wild mammals. Despite its many appeals, this method requires researchers to hand-measure (i) the pixel distance between the parallel laser spots (inter-laser distance) to produce a scale within the image, and (ii) the pixel distance between the study subject’s body landmarks (inter-landmark distance). This manual effort is time-consuming and introduces human error: a researcher measuring the same image twice will rarely return the same values both times (resulting in within-observer error), as is also the case when two researchers measure the same image (resulting in between-observer error). Here, we present two independent methods that automate the inter-laser distance measurement of parallel-laser photogrammetry images. One method uses machine learning and image processing techniques in Python, and the other uses image processing techniques in ImageJ. Both of these methods reduce labor and increase precision without sacrificing accuracy. We first introduce the workflow of the two methods. Then, using two parallel-laser datasets of wild mountain gorilla and wild savannah baboon images, we validate the precision of these two automated methods relative to manual measurements and to each other. We also estimate the reduction of variation in final body size estimates in centimeters when adopting these automated methods, as these methods have no human error. Finally, we highlight the strengths of each method, suggest best practices for adopting either of them, and propose future directions for the automation of parallel-laser photogrammetry data.

Appendix

Skimage method without machine learning

As noted in ‘Schematics and Workflow for the skimage and ImageJ Methods’, the skimage method can also be performed without the pre-processing step of machine learning to mask the gorilla.

Here we compare the results of the skimage method with machine learning (as presented in the manuscript, Table A1: row 1) to the skimage method using only image processing steps on the full image, not using the machine learning masks, ‘skimage without machine learning’ (Table A1: row 2). This was performed on the original dataset of 100 gorilla images. The image processing takes longer without the machine learning pre-processing step (Table A2). The skimage without machine learning method returned measurements for 100% of images; however, in one image the laser spots were mis-identified (not included in Table A1). Using skimage with and without machine learning produces very similar results on the gorilla dataset (Table A2). Thus, using the skimage method without machine learning may be more suitable for researchers without access to the processing power needed to run the machine learning step, or who are unfamiliar with machine-learning algorithms. However, the image processing code was developed for use with photos of dark animals against green backgrounds. Thus, for researchers who are using images that are distinctly different in color or animal species, implementing the machine learning step on their photographs will almost certainly yield better-quality masking than image processing approaches based on color alone.

Table A1 Percent differences [pixel differences] among inter-laser distances measured manually vs. the skimage method with machine learning and manually vs. the skimage without machine learning method on the gorilla dataset; diff = difference

Table A2 Computing time (minutes) required for skimage, skimage without machine learning, and ImageJ methods to process 100 images. For ImageJ images that need to be hand-cropped, this takes approximately 5 additional seconds per image of manual effort. Note the gorilla images have a longer processing time because they have more pixels per image

Table A3 compares the three methods as in Table 1, but it reports the absolute differences instead of the percent differences. Table A4 compares the three methods as in Table 1, but it provides percent differences as a function of inter-beam distance.

Table A3 Absolute percent differences [and absolute pixel differences] among inter-laser distances measured manually vs. the ImageJ method, manually vs. the skimage method and via the ImageJ method vs. the skimage method on both study animal datasets; diff = difference

Table A4 Percent differences [pixel differences] as a function of inter-beam distance (cm); data are from rows 1 and 4 of Table 1, split by inter-beam distance in centimeters, diff = difference