Grassland is a network of open source, computer vision and GIS (Geographic Information Systems) software that turns a 2D camera feed into a 3D map simulation with "live" objects similar to the games SimCity® or Civilization®, but with the ability to rewind time and view their entire history.
By giving a node access to any fixed perspective, 2-D camera feed, it produces 3-D, searchable, simulated re-creation of events in the part of the world it's viewing using OpenStreetMap 3D.
Features of the software:
Introduces a trustless, cryptographic proof-of-work algorithm designed for feed-forward artificial neural networks to ensure the data each node is producing is a result of actual computation from neural network inference on the real world and the fabrication of videos or data and the rewriting of history remain computationally infeasible.
Camera frames aren't stored in the database, just the information gathered from each frame, which includes among other things object identification, object geospatial information, a timestamp, a SHA hash of the frame and a hash chain of the activations of certain hidden layers of the CNN during object detection. The frames themselves are only needed for a little while after for random confirmations and then they can be discarded. It's astronomically improbable to have hashed those hidden activations into the correct digest unless those hidden layers were actually computed, the "proof-of-work".
Uses a 3D version of OpenStreetMap with building extrusion and a modified version of Mapbox
The objects the map displays can be singled out to see identity information, historical paths and locations
The map renders objects as simple low-poly blocks (think Minecraft). Extremely lifelike visual rendering on the map is not necessary, neither could it be achieved at the network's current proof-of-work requirements.
Current computational requirements are low as it's only necessary to detect and track people and cars now but that will increase over time. (During development, I was tracking people and cars 24/7 using my pared down Tensorflow on Serverless AWS Lambda and my AWS Lambda costs were no more than $50 per month)
Every few months, "eons", nodes can the next version of the network's deep learning model. Which will detect, recognize and track more objects and activities with greater certainty in order to discern more and more about those objects and reduce uncertainty. So over time the network becomes an increasingly accurate and harder to fabricate representation of the real world.
Software is currently an alpha version. I've still got some multi-processing bugs to work out in Node Lite and some features haven't been completed yet. You can find installation instructions on the Github repos.
Please, submit questions/problems/discussions in the Issues section of the Github repos. Send me a pull request if you have improvements to make.
Downloads
You can download the latest Grassland eon's object detection model using the following URL's.
For use on low power machines like the Raspberry Pi, this node software version is used in tandem with a serverless lambda function performing object detections. So you get "infinite" horizontal scaling of object detection inference for maximum FPS without having to buy expensive hardware.
Grassland Proof-of-Work Diagram
The results of each object detection on a frame is published along with cryptographic hashes from different stages in the neural network
