Description of movement sensor dataset for dog behavior classification

Area of Science:

• Animal Behavior
• Machine Learning
• Sensor Technology

Background:

• Understanding canine behavior is crucial for welfare and training.
• Objective data collection methods are needed to supplement observational studies.
• Previous research has explored sensor-based behavior analysis in animals.

Purpose of the Study:

• To collect and annotate movement sensor data for seven distinct dog behaviors.
• To provide a dataset for training and validating machine learning models for dog behavior classification.
• To develop and share signal processing and classification algorithms for canine movement analysis.

Main Methods:

• Collected six degree-of-freedom movement sensor data from 45 middle to large dogs performing seven behaviors (sitting, standing, lying, trotting, walking, playing, searching).
• Utilized collar and harness-mounted sensors, with data synchronized to video annotations at one-second resolution.
• Repeated data collection for 17 dogs to ensure reliability, varying task order to minimize bias.

Main Results:

• Successfully collected and annotated synchronized sensor and video data for seven dog behaviors.
• Developed and validated signal processing and machine learning algorithms capable of classifying these behaviors.
• The dataset and algorithms are made available for further research in canine behavior classification.

Conclusions:

• Movement sensor data, when accurately annotated, is effective for classifying distinct dog behaviors.
• The provided dataset and algorithms facilitate advancements in automated canine behavior analysis.
• This work supports the development of objective tools for understanding and monitoring dog behavior.