Argos: A toolkit for tracking multiple animals in complex visual environments.

Area of Science:

• Computer Vision
• Animal Behavior Analysis
• Machine Learning

Background:

• Automated multi-animal tracking is crucial for behavioral studies but current computer vision tools struggle with natural environments and motion complexities.
• Existing algorithms often fail in inhomogeneous lighting and cannot handle sudden animal movements, necessitating manual correction and limiting research scope.

Purpose of the Study:

• To introduce Argos, a software toolkit designed for robust multi-animal tracking in challenging, inhomogeneous environments.
• To provide tools for video compression, convolutional neural network (CNN) training, automated tracking, and manual track correction.

Main Methods:

• Argos utilizes a CNN for animal detection and incorporates specialized algorithms for multi-object tracking.
• The toolkit includes graphical user interfaces for efficient training set generation and manual review/correction of animal tracks.
• Video compression based on animal movement is implemented to reduce data storage and analysis load.

Main Results:

• Argos successfully tracks multiple markerless animals in inhomogeneous environments over extended periods.
• The software demonstrates reduced data storage needs, accelerated analysis, and improved handling of difficult tracking scenarios.
• Argos facilitates analysis under conditions previously challenging for automated systems.

Conclusions:

• Argos offers a versatile solution for multi-animal tracking, adaptable to various recording conditions and computational resources.
• The toolkit enhances the efficiency and accuracy of behavioral analysis by overcoming limitations of previous automated tracking methods.
• Argos enables long-term, high-throughput recording and analysis of animal movements in complex, naturalistic settings.