For large-format portraiture to work, using a sensor-stitch method, I believe the total “exposure” time must be under 10 seconds. The longer the capture takes, the higher the risk that the subject will blink, or move in a way that ruins the stitched composite (like an arm sliced off). If it’s windy, one has to worry about hair, trees, etc. In these posts I chronicle my efforts to create efficient digital-stitching large format photography.
Here’s an amusing capture of Alice with some “wheels” on the left in search of a car.
To cover 4×5 inches with a full-frame camera, I need to take 20 to 30 shots with a full-frame sensor (36mm x 24mm), depending on how much over-lap (for the stitching software) I want needs. In my first attempts, I could do this manually in about 30 seconds. In my current version, with an electronic firing mechanism, I am getting below 15 seconds.
A couple of weeks ago, I thought I had a solution. A trick would be to use 2 to 5 cameras, staggered in columns, each firing as they get over one of the partial image areas.
This is crudely how the device would look. To get the sensors as close together as possible, the cameras are mounted bottom to bottom. Unfortunately, for the A7, it’s 25mm from the bottom of the sensor to the bottom of the camera. For both cameras, bottom to bottom, that’s 50mm, and we need to cover at least 18mm of the 24mm height of the sensor during each row scan. That would mean at least 3 scans, horizontal movements, to fill in the space between each camera’s sensor. Although this isn’t practical for 4×5, it could work well for a 6×7 image, roughly 150mm x 175mm; that is, with two cameras one could create a 6×7 image as fast as they could create a 4×5 with one camera.
Anyway, I figured for 4×5 I’d have to push the cameras apart, horizontally, until I could line up the bottoms of each sensor for an 18mm gap. In that way, I can move the cameras up one row at a time.
Confusing as that is, it should hopefully make sense with this diagram. In this scenario, we could alternate/capture two rows at a time.
Each exposure would take under half a second. With two cameras, you’d have to make 4 passes. The first two would cover two rows. Unfortunately, the third and forth would duplicate rows with the bottom camera. In any case, with 2 back and forth movements of the digital camera back, I could capture 4×5 worth of image area.
With an investment in five cameras, each mounted on a row, I could capture the entire 4×5 image in one pass.
Unfortunately, I didn’t think my scheme all the way through. Staggered cameras mean more shots per column. There’s probably a marginal difference between sliding the camera horizontally for a shot and moving a camera up to another row.
There are two components to the time it takes to tile images. The first is the row/column the shot is fired on, the second is the shot itself. In the staggered camera approach, I only factored in the saved time when two cameras capture simultaneously over the image area. I didn’t factor in the time it takes to fire the shots of the staggered camera entering, or leaving the image area. Even with five cameras staggered above each other, I would still need around over 20 columns of shots captured.
I’ve abandoned the staggered camera approach for now.
Pressing the shutter button takes thought and time, funny as that sounds. To quicken the process, I’ve been developing a linear multi-contact switch. For the positive lead I’m using a bearing soldered onto a wire. For the negative end I’m trying various approaches, from wrapped copper wire to staples, as you can see here. For each camera sensor size, the contacts would be position to fire it at whatever distance I want between frames.
When I finally got the electronic triggering mechanism working passably I discovered that the manual movement, and slow responsiveness of the cameras, produced frames that were not consistently spaced. One frame might be too far to the left or right. From that I discovered that most stitching software struggles if the images are off. There is no magic software for stitching poorly aligned images together, even if they contain enough visual data.
Just as I was about to quit, the 10th iteration began to yield results. By using a round bearing, that falls into curved spaces, I was able to feel when I needed to pause the camera movement and it took care of the shutter firing.
Also, I learned that I can get a quicker single-shot rate by setting the camera on continuous mode, even though I’m only taking a single shot at a time. It seems the camera shuts down all image processing to the bare minimum and records images as fast as possible. I learned this about the EOS-M and it seems to apply to the Sony cameras too. However, recently I’ve found I can get into a rhythm with single-shots and don’t use continuous.
Here’s a portrait I did of my friend Mark using the new automated firing mechanism.
Lately, stitching software seems to have more problems with many of my capture sets than it should. I suspect it is from the poor alignment and fit of the tiling mechanism. I have to apply some pressure to the A7 to keep it against the Graflex. I’m probably moving it a smidgen, here and there. As I post this, I built another device and it is working much better.
You can view/download large version of my latest images at Flickr/maxotics.
More later…