Company: The Pixel Farm
Pros: Relatively affordable tracking solution for broadcast, auto-tracking with good post-tracking cleaning and refinement tools.
Cons: Doesn’t seem to allow you to add custom tracking points. Interface can be a little glitchy, and commands from the higher-end PFTrack are left in but greyed out, which can cause confusion.
The market for match-moving software is becoming increasingly crowded – and now there’s another product to add to the list. The Pixel Farm’s PFMatch joins the likes of 2d3’s Boujou, RealViz’s MatchMover, and a handful of plug-in-based solutions in the match-moving market. PFMatch aims to offer a sophisticated toolset and autotracking for the extraction of 3D camera data from film or video footage.
The Pixel Farm actually offers two matchmoving applications – PFTrack and PFMatch. PFTrack is the high-end application, with an accompanying high-end price tag. PFMatch is a cut-down version, offering much of the functionality but with certain restrictions. Priced at £750 makes it a more affordable matchmoving package, suitable for smaller companies and video-artists.
PFTrack is clearly aimed at the higher-end of the market, as it supports unlimited image resolution. PFMatch is restricted to broadcast-resolution footage. Only PAL and NTSC formats are supported, so if you need to work with high-res film then you have to fork out for PFTrack. Match limits you to a single sequence per project.
The interface is dark and pro-looking, and well-laid out. To the left is a panel containing the various parameters and items used in the project, while in the centre there is a main viewing area. Beneath that is the timeline. A toolbar rests at the top of the interface, and the ergonomics of this set-up are quite good.
The panels are divided by bars, which you can drag to re-proportion the interface depending on what you need to see most of at the time. New 2D or 3D viewing windows can be added and titled to make various different layouts, though the program doesn’t seem to make use of Quartz under OS X, so windows redraw clunkily when dragged about.
The toolset is powerful, and the workflow is well-designed. Footage is imported into PFMatch where it is displayed in the main viewing pane and entered below the Footage section of the item panel on the left. You can set various options by right-clicking the Footage item and choosing Edit Format. Here you can remove interlacing from video footage that might interfere with the tracking process.
You can improve image quality by using the built-in image controls in the Manipulate panel. It has noise reduction (a bit like Photoshop’s median blur), blurring/sharpening and brightness/contrast sliders. These filters are quite slow and their output isn’t cached, so it would be better to make these adjustments in Photoshop (batch processing a sequence of images with an Action), or a compositing tool such as After Effects or Combustion.
There is a simple method for compensating for lens distortion. In Distortion mode, you draw a straight line along a feature in the image which is supposed to be straight. By shift-clicking over the line you place subdivision points, and PFMatch pops up a zoom window so you can accurately match the line to the feature. The program then calculates the distortion and displays a blue distortion grid over the image. You can export the corrected footage.
Tracking and 3D calibration can then be performed. You can track first and then calibrate, or do it all on one click. This takes time, but PFMatch is quick considering it tracks using the full colour method, which is more accurate than luminance only. The camera extraction can be checked by opening up a 3D view and rotating, and by placing a 3D object in the scene and playing back the footage. If the object appears to stay locked in its spot then it’s a good track – if it starts slipping and sliding then you need to make some adjustments.
Back on track
This is where a good tracker proves its worth. As handy as autotracking is, there are always shots that will need to be done by hand. In PFMatch, you can add extra user-defined tracking points to help improve the solution, and set various parameters for the tracking process (such as to enable/disable certain colour channels, pixel similarity mode, and tracking area). Unfortunately, the calibration process does not use them.
The option to use User Features is greyed-out in PFMatch, this feature is only available in the more-expensive PFTrack. A cruel trick – surely it would be less confusing to simply gray out the New User Feature menu item. Marking pairs of parallel lines in the footage helps the program determine the focal length of the lens, which plays a big role in improving the quality of the calibration.
Feature Cleaning lets you remove low-quality tracking points using an interactive threshold slider. Once set, you can re-calibrate the scene to improve the output. Calibration can operate in two different modes if the solution is not working out, but there’s also a Bundle Adjust feature for tweaking the calibration process if you find the solution isn’t quite right.
After some adjustment you can get a decent track from just about any sequence, and PFTrack is straightforward to make the fine-tuning adjustments. Once the tracking is complete you can export the 2D and 3D data to a large variety of formats including Maya .ma, LightWave LWS, Max Script, and various Shake formats, as well as scene data to PF Barn, the company’s image-based modelling application. PFMatch is a quality application that brings quality motion-tracking to a wider user base.