Here is my final hand held shot. There is not much to it in terms of CG material, however it did prove quite a challenge to track. The compositing was aslo quite time consuming due to the shere amount masking that was required. Although in general I am pleased with the final result. 

Here is the raw footage from my hand held shot. I gave it quite a bit of movement to put PF Track through its paces.

In order to place the persons reflection onto the sphere I duplicated the masked layer and reduced its size by about 80%, then once again I had to manually match it to the movement of the sphere, always ensuring that the reflection remained in the centre. I then applied a Gaussian blur to the layer and colour corrected it to blend it into the surface of the sphere. Finally I used the optics compensation tool to warp the layer so that the reflection matched spheres convex face. 

For the second part of my handheld shot I have someone being chased by a large reflective sphere. I decided to do this because it would result in me having to mask out a moving person frame by frame in order to place a 3d object behind them. Also seeing as the sphere is reflective I would have to somehow add the person’s reflection onto its surface. To start with in PF Track I used a basic mask to exclude the moving person as failure to do so would have resulted with inconsistent tracking points. Once again I had to manually track about 20 frames of footage as the camera does shake quite severely.

The reflective sphere was animated and rendered in a similar fashion to the one in the previous shot. Once the image sequence was in After Effects I had to apply some colour correction using Levels as well as Curves to change the individual properties of the RGB values. Saving my footage as .EXR also allowed me to match the exposure of the sphere to that of my environment.

In order to place the sphere behind the person I had to mask him out frame by frame (70 in total, about 2 days worth of masking!!). Seeing as I would be duplicating the running person in the reflection of the Sphere I had to mask out the entire person. For example in most of the shot his legs don’t obstruct the sphere and therefore masking them would have been pointless, however seeing as I need the full person in the reflection I had to mask everything.

I animated the sphere using a spline to determine its path and then some Max Script to make it roll. I then used the Vray material wrapper and applied it to every object in the scene except the reflective sphere. The VRay material wrapper allowed me to make the objects it was applied to transparent in the alpha channel of the final image sequence. It also allowed me to preserve the alpha of the shadow cast by the reflective sphere. Even though the 3d objects in the scene are not visible in the final render, their reflections are still visible on the surface of the reflective sphere.

The first part of my match move has a chrome ball falling from the sky and colliding with the pavement (with some help from reactor). I used my custom HDR map as an environment map within the scene, giving the steel ball reflections relevant to the environment in the footage. In Photoshop I created a cracked pavement texture as well as an opacity map, allowing the crack to blend into the footage. This appears when the steel sphere collides with the ground. I created a plane and applied the cracked surface texture to it as well as the opacity map to ensure that only the cracks were visible. I then placed the plane within my tracked scene and rendered the cracked pavement in a separate pass, allowing me to apply colour correction to it separately.

For my final shot I’ll be concentrating more on the actual track and less on the 3d objects within the scene. I’ve recently recorded some very shaky footage just to test the limits of PF Track. I filmed the footage with a shutter speed of 1000/1 to ensure that the footage does not blur when the camera moves, thus giving PF Track sharp footage to track. My focal length was 4.2mm which resulted in some severe lens distortion which I corrected using the same techniques I implemented in my nodal pan (lens distortion grid, optics compensation ect). Whilst tracking my footage in PF Track I encountered a few major problems. To start with PF Track had trouble tracking about 1/3 of my footage, this happened pretty much every time I rotated the camera. To make matters worse it also happened in random intervals throughout the footage. In an attempt to fix this I increased my target features (located in tracking parameters) from 500 to 3000, I also changed the threshold from 0.01 to 0.005 to try and get PF Track to recognise smaller points within the footage. Neither of these worked, I even had a go with PF Track version 5.0 and had minimal success. This resulted in me having to manually track about 70 frames using my scale 3d scene as reference against the footage. In the end I did prevail however it did take 2 days!

My final match move contains some highly reflective 3d objects and so it was important to create a High Dynamic Range environment map captured in the same area as where I filmed my handheld shot. To do this I photographed a chrome ball bearing at different exposures (usually at 1 stop intervals). I then assembled a HDR image in HDR Shop 2.0 based on the images I had captured. I was then able to convert the steel bearing into a latitude/longitude image, showing 180 degrees of the environment. A 360 degree image was not necessary as I generally don’t rotate the camera more than 180 degrees in my footage.

Finally here is a simple breakdown of my workflow regarding this shot.

Camera: Panasonic HVX

Focal Length: 7.5

Resolution/FPS: 1920x1080 25FPS

Camera distance from Pillars: 1530cm

Driveway width: 500cm

Pillar height: 200cm, Pillar Width: 56cm

Distance between Pillars: 427cm

Lens distortion was corrected in After Effects CS4 using the optics compensation tool. The footage was then tracked in PF Track, taking care to mask out moving objects such as trees. The final track was exported to 3DS Max using MaxScript. Scene geometry was modelled and matched with the camera perspective. During the tracking process PF Track recorded points along the top of the gate which aided in matching the footage with the geometry. Additional modelling was done in ZBrush. The final sequence was rendered at 25fps to match the original footage. All elements were then composited in After Effects and exported.

I’ve decided to add a few more additions to my original track footage just to make it more interesting. Seeing as I could not think of anything interesting to do with the driveway I just removed it completely, replacing it with a muddy trench. In a way I’m faking depth in the scene by adding a trench that’s not actually there. The new depth added to the scene helps reinforce the perspective of the track, plus the 3d trench would suffer from parallax if the exported camera was not properly matched in PF Track.

The trench was made in ZBrush and was based upon a basic mesh which I modelled in 3DS Max using my basic checker scene as reference for scale. It was then composited in After Effects.