Ethoflow

Ethoflow automates quantitative analysis of animal behavior using computer vision and artificial intelligence to detect, segment, track, and classify behaviors such as kinematic patterns and trophallaxis in heterogeneous environments.

Key Features:

• Object detection and instance segmentation: Implements object detection based on instance segmentation to detect and delineate individuals in heterogeneous field and laboratory environments.
• Multi-individual tracking with identity maintenance: Tracks multiple individuals simultaneously while maintaining individual identities for analyses of group dynamics.
• Convolutional Neural Networks (CNNs): Employs CNNs to classify and evaluate complex behaviors beyond simple positional tracking.
• Automatic training data generation: Generates training data automatically using heuristics to train AI models for detection and behavior assessment.
• Modular AI processing and computational scaling: Separates AI processing modules to enable kinematic measurements on standard CPUs and complex behavior assessments on machines equipped with GPUs.
• Kinematic measurements: Performs quantitative kinematic measurements of movement patterns and physical interactions.

Scientific Applications:

• Kinematic assessment: Enables quantitative analysis of movement patterns and physical interactions among animals.
• Social behavior detection (trophallaxis): Detects trophallaxis in bees to investigate social interactions within groups.
• Group dynamics analysis: Facilitates studies of group behavior by tracking multiple individuals and preserving identities over time.

Methodology:

Uses instance segmentation-based object detection, convolutional neural networks for behavior classification, heuristics to automatically generate training data, multi-individual tracking with identity maintenance, and a modular architecture that separates AI processing to perform kinematic measurements on standard computers and complex behavior assessments on GPUs.