Asimov is a Python package and interactive real-time vision system that can also be run as a standalone program. It processes camera input via OpenCV, recognizes hand gestures using cvzone (which wraps MediaPipe), and provides configurable visual effects and “glitches.” In main_virtualcam.py, cvzone is also used for face detection to implement a “privacy eyes” overlay. Asimov includes utility functions for distance checks, logging, and safe path computation, making it suitable for building interactive applications or demos.

Usage modes: Asimov can either be imported as a Python package in your own scripts, or run directly using python -m asimov, which will execute a pluggable main loop such as main_virtualcam.py. This allows you to quickly launch demos with gesture recognition, glitch effects, and virtual camera output.

Virtual camera support: Compatible with OBS (tested) and Zoom (tested). meet.jit.si (Jitsi) does not currently work (black preview), and Discord shows the virtual camera in the dropdown but displays a black frame.

Legacy demos: GTK-based demos exist but are no longer actively maintained. OpenGL-based demos exist but are not used in production.

• Real-time camera input using OpenCV's VideoCapture
• Hand and gesture recognition via cvzone (which uses MediaPipe) and OpenCV
• Face detection via cvzone for the “privacy eyes” overlay in main_virtualcam.py
• Gesture state management with debounce logic and configurable thresholds (gestures.py)
• Demo code uses gestures.py to detect pinch and spread gestures, triggering visual effects or interactive responses
• Configurable “glitch” and distortion effects with timing and randomization
• Utility functions centralized in asimov.common:
  • logentry() – logging events and actions
  • safe_distance() – distance checks for interactive safety
  • get_safe_path() – computing paths that avoid collisions or restricted zones
• Extensive use of named constants to avoid magic numbers and improve readability
• Can be imported as a Python package or run directly as a program
• Supports pluggable main loops via Python modules (e.g., main_virtualcam.py) for demos and glitch effects
• OpenGL-based demos exist but are not used in production
• Virtual camera output via pyvirtualcam, compatible with OBS (tested) and Zoom (tested); meet.jit.si (Jitsi) does not work (black preview); Discord shows the virtual camera in the dropdown but displays a black frame
• Legacy GTK-based demos for video display and interaction (untested in recent versions)

Asimov depends on the following Python packages:

• pyvirtualcam — Sends frames to a virtual camera.
• opencv-python — Python bindings for OpenCV for computer vision tasks.
• numpy — Fundamental package for scientific computing.
• mediapipe — Cross-platform framework for building multimodal applied ML pipelines.
• PyQt5 — Python bindings for the Qt GUI framework.

And the following system packages should be installed via dnf on Fedora:

• ffmpeg — for video encoding and decoding (optional).
• gstreamer1-plugins-base — base GStreamer plugins (optional).
• gstreamer1-plugins-good — additional “good” GStreamer plugins (optional).
• mesa-libGL — OpenGL library for rendering.
• qt5-qtbase — Qt5 base libraries for GUI support.
• opencv — OpenCV system libraries.

Please make sure all Python and required system dependencies are installed before running Asimov.