<scp>AnimalTA</scp> : A highly flexible and easy‐to‐use program for tracking and analysing animal movement in different environments

Tracking the movement of animals is one most important methods used by scientists to study animal behaviour. Before arrival modern technologies, researchers measured locomotion simply observing and describing it in objective way possible (Kline, 1899). The use cameras computers brought new tools researchers, allowing an easier, faster more measuring A growing number video tracking programs allow automated analyses behaviour (Dell et al., 2014), power precision these are rapidly improving. These state-of-the-art technology, such as artificial intelligence, follow up a hundred identifiable individuals same arena (Ray & Stopfer, 2022; Romero-Ferrero 2019; Walter Couzin, 2021). have their pros cons, despite users, some practical difficulties remain unsolved. Some limitations existing discourage research. first those accessibility: good portion do not come with simple installer. Instead, user commonly requested pre-installed additional software, MatLab, R or Python environment (Harmer Thomas, Panadeiro 2021; Sridhar 2019). Occasionally, also required write codes modify ones (Mathis 2018; In recent paper, al. (2021) reviewed 28 different programs, among which only four were user-friendly (Panadeiro Many users often get frustrated when they asked install supplementary need continuously look for information user-unfriendly guidelines. Although efforts might seem negligible people, dissuasive implies significant loss time (personal observations). second problem that requirements quality recording conditions too demanding. For instance, perform over whole duration (e.g. Pérez-Escudero 2014; 2021), so crop videos before analysis using editing program. Most demand specific format, from output user's device efficient tracking, record under strictly controlled conditions. example, our knowledge, none tolerant camera tremors. stop case tremors, continue but unable correct target's trajectories biased tremors Rodriguez high resolution (in both spatial temporal scales) lighting homogeneous background environments (see, e.g. guidelines ToxTrack idTracker: 2018). Meeting may pose challenge, particularly laboratories limited budgets. Moreover, many cases, difficult meet video-recording conditions, necessary complex (and realistic) sake design natural Finally, lack ability simultaneously analyse multiple arenas within single Because replicate experiment times, tests several arenas. They experimental settings, Here, we introduce AnimalTA, software whose aim bring provide easy-to-use tool will numerous arenas, recorded realistic this end, AnimalTA provides built-in light correction, image stabilization perspective allows fix parameters related target detection filtering. program manually errors, re-run selected part easily obtain ready-to-use data adapted needs. To demonstrate efficiency compared performances two other analysing various installer can be downloaded at: http://vchiara.eu/index.php/animalta, source code accessible https://github.com/VioletteChiara/AnimalTA (Chiara, 2023). This has been developed Windows OS uses libraries, mainly Tkinter Graphical User Interface (GUI), OpenCv2 treatment decord importation reading. .exe file (built Inno Setup: https://jrsoftware.org/isinfo.php). reading them mandatory learn how AnimalTA. comprehensible panel always visible on top right corner application guide advise about what actions can/must made. available five languages (English, Chinese, Spanish, French Galician). work at time, although maximum depend characteristics computer videos. accept all sizes resolutions if targets occupy less than one-third image. We each occupies 15 pixels. recommend pilot test experiments whether performs well able complete projects thousand 3-min (resolution: 1280 × 720) without any ordinary laptop. .avi files formats, convert adding project (tested avi, flv, mkv, mpeg, mpg, mp4, mts, m2v, m4v wmv). imported raw modified prepared merged cropped, scale defined frame rate changed. It apply filter unstable recording. identify video, approaches. By default, found adaptive thresholding method objects darker local identified. frequently (but see 2019), producing better results subtraction method, especially changing. choose automatically creates (an targets). created calculating median value pixel subsample images grayscale video. fail remains place half However, visualize produced its errors starting process. freely preferred option between method. save material resources, behavioural separate simultaneous individual IdTracker: TRex: Mathis DeepLabCut: few define ToxTrack). multi-arenas management, trajectory associated identity confusion cannot occur (see Supplementary Video B illustration: while did confuse targets, TRex made errors). With track either shapes, circular, elliptical, polygonal even irregular forms. Once prepared, proceed next panel. panel, set optimize tracking. threshold values size), specify arena, limit distance move frames, activate flickering correction add erosion dilation filters binary During process, effects parameter settings immediately visualized flexible ensure accurate low-quality changing luminosity imperfections. If identified frame, identities assigned minimize last seen position current (Trucco Plakas, 2006). When contact, identification produced. reduce KMeans function scipy library find centre (Sridhar performed operation applying preparation, processes similar wishes manner, needs once. After videos, altering results, necessary, via methods: (i) coordinates dragging dropping it, (ii) removing previous click positions, (iii) ‘interpolation’ select frames interpolated (iv) ‘redo tracking’ redo process parameters. Savitzky–Golay smoothing (Press Teukolsky, 1990) applied noise imprecisions. digital applies convolution data: successive subsets low-degree polynomial linear least squares way, smoothed, accuracy improved being distorted. functions obtained Movement target, total moved, average speed, meander value, others, analysed following brief straightforward instructions shown analyses, parameters, considered moving, calculate proportion explored area, shape cell size grids measure exploration (if solution chosen user) area around moving. instantly shows changes affect results. elements interest, spent user, border mean point, others. examples illustrated described Figure 1. present inside obtains inter-individual interactions, between-target close measures. rename after Thus, interpret looking 2f). exporting tracked visualizes showing step preparation black white, erosion) identities; 1). facilitate review illustration strengths ultimate nature, flexibility quality, simultaneously, integration basic potential errors. limitations. cross central positions overlap. distribute three-dimensional (3D) environment, fish swimming water Video: E). two-dimensional (2D) other, preserved make contacts Videos: M, O). Therefore, who grouped animals, flying strong competence distinguishing individuals, idTracker (Pérez-Escudero idTracker.ai (Romero-Ferrero 2019) (Rodriguez Another limitation portability, currently operating system. does include intelligence (AI) algorithms. recognize distinguish postures body, head legs. contrary AI-based (Feng Xiao, Ray 2022), properly moving environments. tested 19 kinds (slime-mould cells, ants, spiders, crickets, snails, fish, mice, sheep, primates vacuum robot; Videos S1–S19), per (Figure 2, Table 1, S1–S19, S1). comparison purposes, programs: 2018) (Walter chose because objectives features emphasis performing fast analyses. multi-arena multi-individual requires backgrounds quality. specialized individuals. estimate three ground truth data. aim, body ImageJ (Schneider 2012); Supporting Information, Data S1. then resulting (Table counted times swapped Note TRex, advice default range change comparisons summarized presented higher rates including had (i.e. I, K, L). was generally negligible, suggesting except (video S). correctly estimated far data, 3D Nevertheless, ToxTrack, As expected, maintained during separated B, showed comparable performance common D, E, H, P, Q confused D E), swaps shoaling making 30% fewer swaps. surprising because, algorithm crossing. Better precision, tolerance low article, highly movement. already programs. prominent accessibility, major poor capacity maintain environment. useful perfect due nature species, budget. Violette Chiara coded Sin-Yeon Kim led writing manuscript. supervised work. Both authors contributed critically drafts gave final approval publication. grateful Iago Sanmartín Villar, Srikrishna Narasimhan, Mireille Chiara-Thedy, Jean-Yves Qunli Yao help translating testing thank José Noguera Alberto Velando helpful discussion anonymous reviewers constructive comments Richard Bon, Enikö Csatka, Josselin Duffrene, Audrey Dussutour, Pascal Girard, Adrien Lamaze, Noguera, Benjamin Portal, Gema Trigos Peral, Villar Piotr Ślipiński sharing supported research grants provided Ministerio de Ciencia, Innovación y Universidades (RYC-2015-18317; PGC2018-095412-B386 I00). V.C. funded Fyssen Foundation grant. All declare no conflicts interest. peer history article https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/2041-210X.14115. https://zenodo.org/badge/latestdoi/486264498 along (Data S1) Tables S1–S3 https://figshare.com/projects/AnimalTA/158423. S1 S3. figshare. Dataset. https://doi.org/10.6084/m9.figshare.21977324.v2. Original S1–S19. Media. https://doi.org/10.6084/m9.figshare.21977267.v2. Tracked videos: S20–S38. https://doi.org/10.6084/m9.figshare.21977294.v2. https://doi.org/10.6084/m9.figshare.21977351.v2. S1: S2: S3: S4: Please note: publisher responsible content functionality supporting supplied authors. Any queries (other missing content) should directed corresponding author article.