CamMatch™ Professional
Professional Camera Matching for Blender
CamMatch™ Professional is a professional camera matching solution developed for accident reconstruction, forensic engineering, and technical visualization. Built as a native Blender add-on and powered by OpenCV, it provides an efficient workflow for camera calibration, camera pose estimation, lens distortion correction, reprojection analysis, and forensic visualization — all inside Blender.

See it in action
Calibration & diagnostics




Capabilities
- Native Blender integration
- Camera calibration
- Camera pose estimation
- Lens distortion calibration
- 2D-to-3D object linking
- Multi-frame camera tracking
- Reprojection error analysis
- Diagnostic overlays
- Monte Carlo uncertainty analysis
- Forensic visualization workflows
What’s included
Your purchase provides access to the product’s repository and official update channel for the stated update term. It is not a proprietary software license; the installed add-on remains free software under the GNU GPL.
- GPL software package for CamMatch™ Professional
- One year of authenticated repository access for downloads
- Repository-delivered updates, bug fixes, improvements, and new releases during the access term
- Repository secret associated with the purchaser email
- Official documentation
- Continued use of already-downloaded versions after repository access expires
Professional Certification (offered separately)
The CamMatch™ Professional Certification Program teaches the complete camera matching workflow inside Blender, including 2D image markers, 3D control points, object linking, camera calibration, camera pose solving, reprojection error analysis, diagnostic visualization, and best practices for forensic visualization — live, virtual, and instructor-led, with practice Blender projects and datasets, and an official CamMatch™ Professional Certificate upon successful completion.
Certification sessions are scheduled periodically by Engineering Dynamics Company based on demand, with class sizes intentionally limited for personalized instruction and interactive Q&A. Contact info@edccorp.com for upcoming session dates.
Built for Professionals
CamMatch™ Professional is designed for accident reconstruction engineers, forensic engineers, engineering consultants, government agencies, law enforcement, universities, researchers, and professional visualization specialists.
OpenCV installs automatically
CamMatch uses OpenCV (the cv2 library) for its computer-vision math, but you don't install it separately. On first run, click Install OpenCV in the CamMatch add-on preferences and it sets itself up — from the copy bundled with the add-on, or from the internet if needed — on Windows, macOS, and Linux, and across every Blender Python version, including Blender 5.1 on Python 3.13.
If a machine is offline or behind a strict corporate firewall that blocks the download, the User Guide's Troubleshooting section has a one-line manual install into Blender's own Python. After installing, restart Blender and click Diagnostics in the CamMatch panel.
Requirements
- Blender 5.0 or newer
- OpenCV — installs automatically from the panel on first run (Windows, macOS, and Linux)
Latest version
v2.8.18 · released 2026-11-03 · delivered through the EDC Software repository with automatic update notifications in Blender.
What’s new
Added
- Copy Calibration From Clip. A fixed camera gives one view, and its lens is often calibrated on another clip from the same camera -- a still, or a walk past the targets. Typing the lens in kept the values but not their uncertainty: every estimate held the lens exact. The button (Solve the Camera, beside Reset Intrinsics, shown when the file has more than one clip) copies the calibrated lens, the views it was fitted to and its uncertainty from a clip of the same size, and locks it. Pose, point and distance estimates then re-fit the lens in each trial as they would on the source clip. The panel and the Lens Report say which clip the calibration was made on, and the residuals shown are that clip's.
- Generate Reports saves every report as PDF. The reports were HTML only, and a PDF meant opening each in a browser and printing it. With Save Reports as PDF (on by default) each report, All Reports first, is also written as PDF, printed by Chrome or Edge, whichever is installed (Edge comes with Windows), so it looks exactly like the HTML. Without either, All Reports opens in the browser with the print dialog up. The status line says how many were saved and by which browser.
- Image point in the Point Uncertainty report. Each point's section named the track and the frame it was raycast from, but not where the tracker sat in the picture, so a reported position could not be traced back to the image without the .blend. It now gives the image point the ray went through, in pixels with the origin top-left (the convention Track Details uses). Points estimated before this take it from the Empty, which was stamped with it when it was raycast.
- Pixel Aspect in Camera Parameters. The pixel shape the solve uses was set only by a calibration with Lock Aspect off, and could not be seen or typed. A lens typed in from another file was therefore always solved with square pixels: on D1 NTSC footage (704x480 shown as 4:3, pixels 10 wide by 11 tall) that threw out four of thirteen good targets and put the camera 7.6 in from the truth. Pixel Aspect X : Y now shows the stored shape, with fx and fy in pixels beneath it, and typing it sets what the next solve and calibration use.
- Sort the link list worst first. Sort had one Error order, lowest first, so with more links than the list shows the tracks worth checking sat below the visible rows. It now offers Error, best first and Error, worst first.
- Calibrate (Robust) runs a plain calibration first. Robust picks the points that agree with one camera pose, judged with a lens. It used the lens on the camera as it stood, so on a fresh file, or with a poor starting focal, good points looked bad and were thrown out with the blunder (13 of 16 kept instead of 15; from a 50 mm guess on a 24 mm lens, 6 of 16 at a misleading 0.05 px). It now first fits a plain lens to every point and votes with that as well as the current lens, keeping whichever agrees with more points; the message names the lens that won. Robust multi-frame does the same on each frame. Calibrating before Robust is no longer needed.
- Robust multi-frame offers Set Excluded Tracks to Not Used. After a single-frame Robust calibration the panel warned that the excluded tracks were still used by the solves and offered to set them Not Used; after a Robust multi-frame run it said nothing. A track Robust left out of every frame it was in (a survey blunder, such as trk_11 on all eight frames of the moving-camera exercise) now gets the same warning and button. A track left out of only some frames is not offered, since it fitted the others; the panel names the frames instead, so it can be checked and unticked there.
- Multi-Frame Solve. The switch that shows the multi-frame solve options was labelled Solve Multi-Frame, the same as the button it reveals. It is now Multi-Frame Solve, matching Multi-Frame Calibration and its Calibrate Multi-Frame button.
Fixed
- No false "error px only on 2 observation(s)" after Robust calibration. A point Robust left out was recorded with its error but not its signed X and Y split, so it read back as a record from an older CamMatch, and the panel warned about it (the row and its copy in the calibration store) after every Robust calibration that excluded a point. It now carries the split, and Track Details shows its dX and dY.
- Diagnostic pictures are no longer saved in the file. The error images and the distortion heatmap, made after every calibration and solve, were packed into the .blend as full-size float images, one per frame: a multi-frame calibration over eight frames added about 300 MB. They now last the session. The Open buttons make each one again when it isn't there, for the frame you are on, from the lens and the stored errors. To keep them with the file, tick Keep in File under the Diagnostics buttons (off by default; about 33 MB a picture): pictures already made are saved too, and unticking it leaves them out of the next save. The lens maps the compositor groups read are always saved.
- The noise figures stay editable with the lens locked. Lock Intrinsics greyed the whole calibration uncertainty block, including 2D Noise, 3D Noise and Samples. Those are the shared figures the pose and point estimates use, and a locked lens is when those are being run; they could only be changed in the Solve and 3D Tracker panels. Now the lock greys only Auto Monte Carlo on Calibrate and Measure Calibration Uncertainty.
- The cyan ring shows on a good solve. The error images draw a ring 3x a point's expected error around where the solve puts it. It was drawn before the 8 px circle marking where the point was tracked, and on a good solve the ring is smaller than that circle, so it was hidden. It is now drawn on top.
- Survey error shared by point, not by point count. The uncertainty run gives the calibration and the pose one survey error, since both read the same surveyed coordinates. It matched them by how many points a view had: every view with the same count took the same offsets, whichever targets they were, and a pose frame with the same count as a calibration view took that view's offsets in that view's order, so a target could carry its neighbour's error. Each surveyed point now has its own offset, matched by its coordinates; a point the calibration never used gets its own draw. Stored re-fitted lenses from earlier versions are re-drawn on the next run.
- A cleared noise field no longer blames "an earlier CamMatch". A stored result beside a blank 2D or 3D Noise field was said to come from an older version. Clearing a field after a run is the ordinary way to get there; the note now gives the figures the run used and says the field is now blank.
- No false "camera has moved" on Estimate Point Uncertainty. The check that a raycast point still belongs to the camera re-shot its stored pixel through a pose solved afresh from the frame's points, not through the camera it was raycast from. On noisy low-resolution footage that fresh solve alone lands a few tenths of a percent of range away, so a fixed camera that had not moved was reported as moved (0.34% of range on the D1 exercise). It now shoots through the camera object itself; a re-solve, another clip's camera or geometry in the way is still caught.
- Multi-frame calibration keeps the pixel shape it fitted. With Lock Aspect off it estimated fx/fy and then stored 1.0, so the next solve went back to square pixels. On D1 test footage it fitted 1.104 at 0.69 px and the solve after it put the camera 7.8 in from the truth; it now stores 1.104 and the solve lands 1.6 in from it.
- A solve whose tracks, survey points or links change afterwards is now flagged. Only a recalibration marked a solve stale. Moving a marker, editing a survey point or switching a link off after solving left the frame's stored errors, uncertainty and reports standing, with no warning, and the solve errors image drew the old residual at the new marker: a track that had slipped 376 px read 0.6x. Each solve now records what it was solved from (schema 6). The Solve Camera panel names what changed on the current frame, Diagnostics warns and titles the image STALE, the reports flag the frame, and Re-solve Stale Frames includes it. Frames solved by an earlier CamMatch cannot be checked, and the reports say so.
Changed
- The Lens report gives the coefficients in both Blender's and OpenCV's conventions. It listed Blender's values alone, labelled k1-k4, p1 and p2, and those were read as OpenCV's. For Brown-Conrady they are not: Blender's P1 and P2 are swapped against OpenCV's and one changes sign (OpenCV p1 = -Blender P2, OpenCV p2 = Blender P1), and Blender's K4 is an r^8 radial term OpenCV's standard model has no slot for. The table now shows each Blender value as in Camera Parameters beside its OpenCV counterpart, and says when a non-zero K4 means the lens cannot be reproduced exactly in OpenCV's model. Polynomial K1-K3 are shown to carry across unchanged.
Fixed
- The status message wraps to the panel's width. It was one label, so a long message lost its middle: a calibration's read "Multi-frame calibration over 8 frame(s): RM... render distort Map UV.", with the RMS and worst frame cut out. It now breaks at sentence ends and wraps, up to ten lines.
- A multi-frame calibration's result stays on screen. After the fit, CamMatch draws an error overlay for each calibrated frame, and each one set the status to "Created ... reprojection marker(s)", so the calibration's own message, with its RMS and worst frame, was replaced before it could be read. The message is now kept through those overlays (and through the per-frame overlays of a camera track), and the auto-created diagnostics are appended to it as intended.
- Add Empty at Click lands on a point cloud. It found the point with Blender's ray cast, which only hits faces, so the ray went through a point cloud, or a Geometry Nodes visual of one, to the ground or mesh behind it (4.2 m off in a test), while the 3D cursor, which reads the viewport's drawn depth, landed on the cloud. It now reads the drawn depth too and uses it when it is clearly nearer than the mesh hit; on a mesh the exact hit is kept. Measured: mesh, instanced cloud and Mesh-to-Points cloud all within 2 cm of the clicked point, and the 3D cursor is left where it was.
- The User Guide said to crop an overscanned render back to the clip size. The CamMatch compositor group's Scale node (Render Size, Stretch) already scales the composite to the overscan size with the clip's framing; the guide now says to scale the result down to the clip's resolution to deliver at the clip's size.
---
Download CamMatch-*.zip below and install in Blender via Edit > Preferences > Add-ons > Install from Disk. Requires Blender 5.0+.
Install & updates
Your repository secret is delivered instantly — it appears on the confirmation page as soon as you check out (and free products issue one after a quick sign-up). Then, in Blender 5.0 or newer:
- Open Edit → Preferences → Get Extensions.
- Open the Repositories dropdown (top right) →
+ → Add Remote Repository, and paste
https://extensions.edccorp.com/index.json. - Enable "Requires Access Token" on that repository and paste the repository secret you received into the Secret field.
- Your products appear under Available — click Install.
- Make sure the add-on is enabled. Installing usually ticks its checkbox for you, but if the product's sidebar tab doesn't appear, open Edit → Preferences → Add-ons, search for the product, and tick its checkbox to activate it. Updates then arrive automatically.
- Save your preferences so this sticks. Blender normally saves them automatically; if it doesn't, open the ≡ menu at the bottom-left of the Preferences window and choose Save Preferences — otherwise you may have to re-enable the add-on next time you open Blender.
Upgrading from an older, manually-installed version of this add-on? You can disable or remove it anytime from Edit → Preferences → Add-ons — search for it, untick it to disable, or use Remove to uninstall. The new extension replaces it.
Because the installed add-on is GPL-licensed free software, copies you have already received remain yours to use, study, modify, and share under the GPL. When the update term ends, repository access for future official builds and automatic updates may require renewal.