Waldo Vision Engine (waldo_vision)
waldo_vision is a multi-layered computer vision engine built from scratch in pure Rust. It is designed to detect, track, and analyze significant events in real-time video streams, acting as a high-performance pre-filter to more expensive AI analysis systems.
This crate was specifically designed to serve as the vision capability filter for the Corpus AI Companion project.
Key Features
- Temporal Analysis: Uses a multi-channel statistical model to learn "normal" environmental behavior and detect anomalous changes.
- Spatial Grouping: Implements a "Heatmap Peak-Finding and Region Growing" algorithm to identify coherent objects in motion.
- Behavioral Analysis: Includes a robust object tracker that adds object permanence, tracking events over time to form complete "Moments."
- High-Level API: Provides a simple, powerful
VisionPipelineAPI for easy integration. - Tunable: All key sensitivity thresholds are exposed in a
PipelineConfigstruct.
Integration
For detailed instructions on how to integrate this crate into your project, please see the INTEGRATION_GUIDE.md.
License
This project is licensed under the MIT License.