Movement: a Python Toolbox for Analysing Motion Tracking Data

Movement: a Python Toolbox for Analysing Motion Tracking Data N Sirmpilatze1,2, CH Lo1, S Miñano1, W Graham3, S Prins3, BD Peri4, S Keshavarzi4, AL Tyson1,2. 1Sainsbury Wellcome Centre, UCL, 2Gatsby Computational Neuroscience Unit, UCL, 3Advanced Research Computing Centre, UCL, 4Department of Physiology, Development and Neuroscience, University of Cambridge. Neuroinformatics Unit neuroinformatics.dev References 1. A. Mathis et al., DeepLabCut: markerless pose estimation of user-defined body parts with deep learning, Nat. Neurosci 21, 2018. 2. T. D. Pereira et al., SLEAP: A deep learning system for multi-animal pose tracking, Nat. Methods 9, 2022. 3. K. Luxem, et al., Open-source tools for behavioral video analysis: Setup, methods, and best practices, eLife 12, 2023. Outlook movement is developed by a dedicated team of engineers alongside a large community of contributors, committed to continually improving and maintaining the package long-term. Our planned features include: • More advanced data cleaning methods, e.g. Kalman filters • Coordinate system transformations, e.g. from allocentric to egocentric • Methods to annotate regions in space and events in time • Support for specialised modules, e.g. gait analysis, social interactions, collective behaviour • Integration with tools for neurophysiological data analysis A unified interface for motion tracks With movement you can import data from your framework of choice as an array with standard dimensions, that you can conveniently access via indices or descriptive labels. movement can accommodate data in 2D or 3D, tracking a predefined set of keypoints or bounding boxes, from single or multiple individuals of any species. Abstract Deep learning-based tools, such as DeepLabCut1 and SLEAP2 now allow researchers to track animal movements in videos with remarkable accuracy, without requiring physical markers3. However, there is still a need for standardised, easy-to-use methods to process the motion tracks generated by these tools. To address this, we are developing movement, a free open-source Python package for analysing motion tracks. A variety of applications movement supports an ever-expanding list of applications, including: • data cleaning—outlier detection, interpolation over gaps, smoothing • visualisation—trajectories, occupancy heatmaps, and regions of interest (ROIs) overlaid on video frames • motion quantification—kinematics, head direction, time spent in ROIs, distances between keypoints and individuals, path length etc. and use our napari-based GUI to interactively view and explore motion tracks over videos. Join the movement Feedback and contributions are welcome! Learn more and get in touch: movement.neuroinformatics.dev github.com/neuroinformatics-unit/movement neuroinformatics.zulipchat.com You can call movement in Python scripts, We plan to leverage napari’s rich annotation features and allow users to define regions of interest over video frames as well as tag specific events of interest in time. Our long-term goal is for the GUI to support all common analysis workflows.