Agentic Counter-Surveillance Analysis

This project is an attempt to create a pipeline with independent AI agents to do counter-surveillance analysis. This multi-agent system, uses publibly available data from OpenStreet maps, to analyze surveillance infrastructure. 

The project has two agents, up until now, has two agents.

• Scraper Agent: Downloads surveillance data from OpenStreetMap via Overpass API
• Analyzer Agent: Enriches data using local LLM analysis and generates visualizations

The analysis can produce:

• Heatmaps, to get an idea of the density of the infrastructure 
• Surveillance hotspots, using a DBSCAN algorithm
• Summary statistics, for camera types, operators, surveillance zones.

Under the hood the project uses a LLM and the system operates completely locally without external APIs focusing on a private and secure manner and the agents have internal memory.

Currently the agents can accessed via a Command Line Interface (CLI).

Next steps, include the development of an intuitive User Interface (UI) and a third path finding agent that will try to find routes that are not heavily surveilled, if any.

For more information on usage, you can read the README file.

Below are some examples of the system's outputs:

The source code of the project is open source and lives over at GitHub:

https://github.com/jethronap/UNDO-agentic

The UNDO Team.

CCTV Detection Algorithm (YOLOV8)

For the past year the project has been planning and implementing workshops with activist communities in Malmö, Helsinki and Copenhagen.

The aims of these workshops are multiple. First and foremost, we try to understand and work around the restricitons set up by institutions, namely the police, in researching ubiquitous surveillance of urban spaces. Then, we try to gather requirements for potential software solutions that:

1. Could be of use to activists communities

2. Could help with research and understanding the state of the surveillance infrastructure deployed 

So, moving towards this direction we have trained and developed an algorithm that can detect CCTV in provided images. This is a small fisrt step and we plan to use this simple detector on a bigger system that can detect, analyse and provide detailed descriptions of the surveillance cameras. 

We trained a YOLO8 model on our custom dataset for 20 epochs. Most images used in training have been collected by us through ethnographic research, some images used in this project come from the dataset of the Fuziih CCTV-Exposure. 

The model so far has the following detection metrics:

• Precision and recall have reached balanced high scores (above 83%).
• mAP@0.5 (object detection quality at 0.5 IoU threshold) reached 87%.
• mAP@0.5–0.95 (stricter localization accuracy) reached 41%.

The code for the detector is open source and lives over at GitHub:

https://github.com/jethronap/cctv-detection-UNDO

A preview of the application that can be tested is hosted at HuggingFace:

https://huggingface.co/spaces/jnap/UNDO-project 

Please contact us if you want the dataset or the weights of the model.

The UNDO team.