WiseNET

WiseNET provides a multi-camera indoor video dataset for human activity monitoring, people detection, and tracking, enriched with Building Information Modeling (IFC) contextual data and manual and automatic annotations.

Key Features:

• Multi-Camera Multi-Space Video Sets: Comprises 11 video sets (62 single videos) recorded by six indoor cameras across multiple spaces, totaling over one hour of footage.
• Rich Contextual Information: Includes Industry Foundation Classes (IFC) files representing Building Information Modeling (BIM) data that specify environment structure, topology, and elements.
• Diverse Human Actions: Captures actions including walking, standing/sitting, remaining motionless, entering/leaving spaces, and group merging or splitting.
• Comprehensive Annotations: Provides manual and automatic annotations with meta-information for people detection and tracking; automatic annotations use detectors ranging from traditional machine learning approaches to deep Convolutional Neural Network (CNN) models.

Scientific Applications:

• Human Activity Recognition: Supports development and evaluation of algorithms for recognizing and interpreting complex indoor human behaviors using video and BIM contextual data.
• Building Management Systems: Enables integration of activity monitoring data for energy efficiency, security, and occupancy-based analytics.
• Computer Vision and Machine Learning: Suitable for training and evaluating people detection, tracking, and activity analysis models.
• Smart Building Technologies: Facilitates research combining BIM and video to enable adaptive building behaviors based on human presence and activities.

Methodology:

Videos were annotated both manually and automatically, with automatic annotations produced using a range of detectors from traditional machine learning approaches to deep Convolutional Neural Network (CNN) models.