Skip to main content

PolyChase

CAMERA TRACKING & MATCHING
📖2 min read

Introduction​

Polychase is a free, open-source add-on for Blender that performs mesh-based 3D motion tracking.

Its core function is to track the movement (pose, or rotation and translation) of a rigid 3D object (mesh) within video footage. It is an alternative to commercial tools like GeoTracker.

The process relies on:

  1. 3D Mesh Input: Requiring a 3D model of the object being tracked.
  2. PnP Solving: Using the Perspective-n-Point (PnP) problem, where 3D points on the mesh are matched to 2D locations in the video.
  3. Optical Flow: Automatically generating correspondences using sparse optical flow data.
  4. User-Guided Correction: Allowing manual correction via a "bin mode" (interactive 3D pinning).
  5. Refinement: Using factor graph optimization (a form of bundle adjustment) to globally smooth and refine the final motion trajectory.

The outcome is a 3D mesh that remains accurately "glued" to the moving object throughout the video sequence.

Key Features​

Smooth Follow

Makes the camera follow a target object with adjustable smoothness and lag.

Damping Control

Control how quickly the camera reacts to the target's movement.

Installation & Usage​

Installation​

  1. Download the addon from https://github.com/theartful/polychase
  2. Install the script/addon via Edit > Preferences > Add-ons.

Usage​

A step-by-step guide for using the Polychase add-on for motion tracking in Blender:

1. Scene Setup and Video Analysis​

This phase prepares the scene and extracts initial motion data from the video.

  1. Import Mesh: Import the 3D mesh of the rigid object intended for tracking.
  2. Create Tracker: Create a new Polychase tracker object.
  3. Define Inputs: Select the appropriate geometry (the mesh), the camera, and the video clip within the tracker settings.
  4. Set Database: Specify a location to save the SQLite database file. This file will store the generated optical flow data.
  5. Analyze Video: Start the video analysis process. The tool automatically generates sparse optical flow data.

2. Initial Mesh Alignment and Masking​

Align the 3D mesh with the object in the footage and define areas to exclude from tracking.

  1. Set Focal Length: Adjust the camera's focal length in the tracker settings if the value is known.
  2. Enter Bin Mode: Activate bin mode to begin manual alignment.
  3. Align Mesh: Create and drag bins (2D points corresponding to 3D mesh locations) to align the imported mesh to the object visible in the video footage.
  4. Mask Geometry (Important Step): Press M to activate masking. Mask parts of the mesh that should not influence the track, such as:
    • Transparent areas (e.g., windows).
    • Moving or deformable parts (e.g., spinning wheels).
    • Inaccurate mesh areas.

3. Automatic Tracking and Inter-frame Correction​

Execute the primary tracking process, pausing to correct drift as needed.

  1. Start Automatic Tracking: Use the automatic tracker function to track forward or backward through the timeline.
  2. Stop and Correct Drift: If the mesh begins to drift away from the object in the footage:
    • Stop the tracking process.
    • Re-enter bin mode to manually correct the object's pose at that specific key frame.
    • Run refinement on the segment tracked so far to adjust the motion between the corrected key frame and the last known good position.
  3. Repeat: Continue alternating between automatic tracking, manual correction, and segment refinement until the desired portion of the video is covered.

4. Final Global Refinement​

Execute a final optimization pass for maximum global consistency.

  1. Refine Entire Segment: Run a comprehensive global refinement on the entire tracked section. This uses factor graph optimization to jointly minimize reprojection error across all frames, ensuring globally consistent motion.

Tutorial Video: https://www.youtube.com/watch?v=W4HNmcjFuLw

Summary​

Camera chase tools add a layer of professionalism to animations by removing the robotic feel of standard parenting. They are essential for dynamic presentations.