Crowdsourced dataset of firefly trajectories obtained by automated stereo calibration of 360-degree cameras
Data files
Jun 12, 2023 version files 47.98 GB
-
README.md
-
s0521uf.zip
-
s0522uf.zip
-
s0523uf.zip
-
s0524uf.zip
-
s0525uf.zip
-
s0526uf.zip
-
s0527uf.zip
-
s0528uf.zip
-
s0601ic.zip
-
s0603ic.zip
-
s0606ic.zip
-
s0608ic.zip
-
s0609ic.zip
-
s0610ic.zip
-
s0611ic.zip
-
s0612ic.zip
-
s0613ic.zip
-
s1601io.zip
-
s1602io.zip
-
s1604io.zip
-
s1605mx.zip
-
s1606mx.zip
-
s1607mx.zip
-
s1609mx.zip
-
s1610mx.zip
-
s1612mx.zip
-
s1613io.zip
-
s1613uf.zip
-
s1614ig.zip
-
s1615uf.zip
-
s1615ux.zip
-
s1617uf.zip
-
s1617ux.zip
-
s1618ux.zip
-
s1619ux.zip
-
s1706ig.zip
-
s1710im.zip
-
s1711im.zip
-
s1723im.zip
-
s1806ik.zip
-
s1809ik.zip
-
s1810ik.zip
-
s1813ik.zip
-
s2523ul.zip
-
s2524ul.zip
-
s2524ur.zip
-
s2528ul.zip
-
s2529ul.zip
-
s2601ur.zip
-
s2607ia.zip
-
s2607ic.zip
-
s2609ic.zip
-
s2609io.zip
-
s2614ya.zip
-
s2615xx.zip
-
s2616ux.zip
-
s2616xx.zip
-
s2617bw.zip
-
s2617xx.zip
-
s2618bw.zip
-
s2618xx.zip
-
s2619bw.zip
-
s2620bw.zip
-
s2620xx.zip
-
s2621bw.zip
-
s2621ug.zip
-
s2621xx.zip
-
s2622xx.zip
-
s2623bw.zip
-
s2623ug.zip
-
s2624bw.zip
-
s2624xx.zip
-
s2625bw.zip
-
s2625xx.zip
-
s2626bw.zip
-
s2626ic.zip
-
s2627bw.zip
-
s2627mx.zip
-
s2627xx.zip
-
s2628bw.zip
-
s2628mx.zip
-
s2629bw.zip
-
s2630bw.zip
-
s2630mx.zip
-
s2630xx.zip
-
s2701xx.zip
-
s2702xx.zip
-
s2705ub.zip
-
s2705xx.zip
-
s2706mx.zip
-
s2706ub.zip
-
s2707xx.zip
-
s2708ip.zip
-
s2708xx.zip
-
s2709xx.zip
-
s2710ip.zip
-
s2711ip.zip
-
s2713mx.zip
-
s2715bw.zip
-
s2716bw.zip
-
s2718bw.zip
-
s2720bw.zip
-
s2724bw.zip
-
s2801ub.zip
-
s2803ub.zip
-
s2809ik.zip
Abstract
Advancements in animal tracking techniques, spanning from migrating mammals to swarming insects, have resulted in remarkable progress in the fields of behavioral ecology and conservation science. Recently, we have devised a method for tracking luminous fireflies in their natural habitat using stereoscopic pairs of 360-degree cameras. This method offers affordability, versatility, and ease of setup; however, the process of camera calibration has remained tedious and time-consuming. Now, we have introduced an enhanced algorithm that achieves spatial and temporal stereo calibration directly from the data, eliminating the need for manual procedures both in the field and during video processing. The algorithm relies on cross-correlation of flashing patterns and numerical estimation of camera pose. Utilizing this improved protocol and processing software, we have compiled an extensive dataset comprising over 100 reconstructed firefly swarms of various species. This data was gathered throughout the United States by numerous contributors following a straightforward protocol. The dataset holds significant potential for advancing our comprehension of firefly collective behavior, facilitating population monitoring, and expanding citizen science initiatives.
Methods
This dataset was obtained by an algorithm for tracking swarming fireflies using pairs of 360-degree cameras. The algorithm performs camera calibration from the data itself so that virtually no user input is necessary besides measuring the distance between cameras. A full description of the algorithm will be submitted to a scientific journal, and sequentially to the bioRxiv. Thanks to this simplified protocol (included here as a supplementary file) and improved software (included here as a code package), we assembled a large dataset comprising over 100 fully 3D reconstructed and time-resolved firefly swarms. This dataset was collected by various contributors across the United States and spans several states and species.
Usage notes
Table 1 provides a summary of the full dataset. Each swarm has been attributed a specific identification (ID) number for convenient referencing. ID numbers are constructed as such: they start with the letter ’s’ (short for “swarm”) to avoid compatibility issues with file names starting with a number; four digits follow indicating the date of recording in YMDD format (first digit is year: 0 for 2020, 1 for 2021, 2 for 2022; second digit is month of the year; two last digits indicate day of the corresponding month); a two-letter code to indicate the (predominant) species recorded.
Tables 1 and 2 provide primary information regarding each dataset, in particular the date, time and location of collection, along with the scaling parameter necessary to convert video units to meters. GPS coordinates of recording sites are provided whenever they seemed appropriate to share publicly, based on considerations of private property, site sensitivity, etc. Tables 3 and 4 provide secondary contextual information, such as predominant species recorded, sunset time (civil twilight), weather conditions, and other brief field notes.
Each swarm, with its specific ID, has a separate zip file, named by as 'ID.zip'. Within each zip file, the main file of interest is a .csv file containing the coordinates of reconstructed flashes. In addition, Matlab structures (ending in ’.mat’) are made available. They contain all the intermediate data used in the course of the processing, notably the input parameters for processing. Finally, a folder called ’info’ contains the metadata for the video files (when available) and a picture of the background in each of the cameras’ field of view (when available). See Usages Notes for a description of the different files.
Usage Notes
.csv files
The xyztkj.csv files contain the coordinates of reconstructed swarms. The first three columns are the xyz coordinates; x and y define the horizontal plane, and z defines the vertical axis oriented upwards. Units are meters. The fourth column represents the corresponding times, in seconds. The fifth column k indicates the streak identity of a flash. The sixth column j indicates the corresponding trajectory.
We emphasize that these coordinates are only partial renderings of the entire swarm. Due to the cameras’ limited light
sensitivity and visual occlusion from the vegetation, only flashes close enough (typically within 10–15 m) can be recorded corresponding to a fraction of swarms which can span several tens of meters
.html files
The .html files can be loaded in a web browser for quick 3D visualization of the reconstructed swarm.
.mat files
The .mat Matlab structure files serve to provide additional processing parameters and intermediate results. All values under the .prm field are input parameters. Values in other fields are the outputs from the different processing steps. These files also allow to re-run the triangulate and trajectorize blocks with different parameters, if necessary (see below).
sphirefly3d software
The software can run from scratch as out = sphirefly3d([s,h]), where s and h are the only necessary inputs corresponding to the real-world spacing and height of the cameras. A window will then open where complementary movies, or folders containing sequences of movies, can be selected. All other parameters can be changed directly from the setprm.m function. Alternatively, an existing sff structure can be used as input, to re-run some parts of the processing: out = sphirefly3d(sff).