First off, I'll analyse why the footage taken a couple weeks ago was not successful, and unusable for the most part.
Then I will go through a second attempt at shooting, and how different problems were dealt with.
Out of the four pieces show in the last post, only one of them was even moderately useful for tracking. This one:
The sideways motion, coupled with the depth of the location resulted in a decent amount of parallax information. However the motion blur present caused the tracking to completely lose the position and pattern every few frames.
Some useful information came out of this regardless of the blurring issues. Patterns tracked effectively on the bricks whenever the camera motions were small and shadows can be used providing the light is constant and the shadows sharp.
The following shots were useless:
The first two were almost tracking along a flat plane, so very little parallax information was available. Plus the combination of the majority of tracking points being a repetitive pattern and the motion blur caused the auto tracking to be impossible, and manual tracking more trouble that it was worth.
The last shot used a camera rotation that was too fast and not really any movement.
Similarly, this shot contained almost no parallax information for the majority of it. And again, motion blur caused what little information there was to become useless.
On the other hand, the above piece worked very well, despite the blur that caused some slipping of tracks, the content and movement created a large number of points to be tracked and the track came out usable with minimal fixing and slipping.
The results of the successes and failings was taken into account and the main issue that needed to be taken into consideration was the motion blur that was destroying all the tracking information.
By default the cameras appear to have a shutter speed of 1/25 second (when shooting at 25fps). So regardless of shooting either indoors or outside in bright sunlight the motion blur was the same. The same amount of movement was being recorded despite the significantly varying light levels.
The only option seemed to be finding the shutter speed settings, which I found on my third shooting attempt. Hidden away somewhere not in the main menu and could only be found by touching the screen.
The first batch of test footage was shot with the shutter speed set at 1/2000 second. The results were somewhat jerky, however a lot of detail was retained in almost all camera movements.
The next step of this will be to find a sweet spot that retains enough detail while allowing at least some motion blur.
Another method tested to control the motion blur was to use the tripod as a counter-weight for the camera, effectively creating a cheap steady cam.
This smoothed out some of the hand held movements, however it had a downside of being extremely cumbersome, and very difficult to both shoot and walk with.
As the steady cam rig training is currently unavailable, my best bet seems to be to either refine this method or use a dolly.
The other issue surrounding the first batches of footage was the narrow field of view present in the cameras. Not only does it make it difficult to get a subject entirely in shot and frame it decently, but it also accentuates the movement of the camera. Which is less than ideal.
Given that no other lenses are available to use, the only options are to A: Shoot something smaller or B: Shoot from further away.
The above fixes were taken into consideration when it was time to shoot some more footage.
The results are as follows:
The bridge footage was successful for the most part. It turns out that grass is a wonderful surface to track as it's full of unique patterns, and if there are lighter/darker patches as well as a few twigs and dead leaves lying there, plenty of contrast too. The bridge itself wasn't too great, as for the most part it's a uniform colour, apart from a few blemishes and the patterns on the corners change shape a little too much. But the path and grass worked very well.
The tree shot worked largely because of the sheer amount of grass present, as well a number of points on the tree, branches and background. All helped hugely by the almost complete lack of motion blur.
I was initially worried about the vehicles moving in the background, especially the lighter ones. MatchMover did attempt to 2D track them, but disregarded them as bad tracks during the 3D solves so they were not an issue. I can imagine that it would be possible for them to move in such a way that they could be seen as parallax information, so that's something to take into consideration for future shoots. The important thing is that things can be moving in shot, just as long as they're moving in the wrong direction.
Building the scene for the bridge composite was simple, I simply used a few points around the path tracks and constructed geometry around them, using the background shader to mask around the edges. The bridge was bright enough to key out with a simple luma key, with a polygon placed in the 3D scene to counter a problem area (a solution that seemed easier and quicker to implement that keying a mask in After Effects). The dust could effect didn't quite come out as intended since I assigned a texture time variable the frame number, rather than the clip time. However, for a test shot it's not worth fixing, especially considering the render time of fluids.
The tree composite required a bit of experimentation in order to recreate the uneven ground, but the technique was quick to set up and successful.
In short, rigid body spheres were placed evenly at the position of trackers on the ground, then a dynamically simulated cloth was dropped on them in order to create a surface that conformed to the shape of the tracking points. I'm sure there's an easier way to do this, but considering how well it worked, I'll stick with the method for now.
Then it was simply a case of spawning a bunch of particles and rendering them out. They were lit using an area light, with the colour sampled from the sky present in shot then rendered out with both that and an ambient occlusion pass. In the colour render the ground used the background shader, and an ambient occlusion shader in the related pass.
Next up, I need to figure out exactly what I'll be shooting for the final piece. Chances are it will involved particles rolling around on the ground.
