After learning about Photogrammetry and the basics of the process, we continue onto the next stage which is to attempt it ourselves and start running through the three main steps. We watched a really cool video that showed one of the best purposes of photogrammetry – its usefulness in film.
It turns out that the technique was actually used quite a bit to build environments for the 2019 remake of The Lion King and some of the animals were also created through scans from zoos and medical data.
After seeing this, we got a quick recap on photogrammetry and the focus of the lesson, capture.
” Photogrammetry is the process of generating 3D model data from an image data set (multiple stacks of images) “
Most of this process occurs in the very first stage where you are capturing image data through photography. Many photos are taken of the subject, from various heights, angles and perspectives in order to capture every conceivable view. We were also given the option of taking a video of the object but although a faster method, it lacks accuracy and is more tedious later on when having to remove motion blur and convert it into an image sequence. Before we could try out either method for ourselves, we went over some more in-depth capture considerations in order to achieve the best result possible.
- Overlap is key – take more photos than you think are needed so that the object is generated as close to the real things as possible
- Shoot in flat, even lighting where possible and in RAW format so that the photos are easier to distinguish by the software
- Ideally, include a random pattern or marker when shooting an object – distinctive squiggles or shapes work best here as they help with alignment
We were also told that if we needed to capture the bottom of an object, we would have to remove any markers and try and photograph the bottom against a plain background.
After going over these tips/requirements, we got into our groups and began planning what we would do with the task at hand. The goal was to capture a data set, so we discussed what we would want as a model and what kind of availability we had for shooting in and around Confetti. This is the plan that we came up with:
– – – – –
Tegan, Will, Jacob, Bilal, Tina
In our group, we will be scanning one of the giant Funko Pop creatures – Grogu. We’re also considering an aesthetic set-up of Will’s present, a box of Red Bull. We were thinking of stacking them on top of and around the box in an advertisement style to make it more interesting than just one can. Additionally, we will also be using Tegan’s face to add some variety (a person among the objects). Someone also suggested Will’s shoe but after seeing Jacob’s, we decide that his would be better since they had more texture because of the mud.
To summarise:
- Grogu
- Red Bull
- Tegan’s face
- Jacob’s shoe
- – – – –
After having settled on what we would be captured, we decided to split the tasks up and have each person take pictures of one of the items. I wanted to do Grogu so I went first and after fumbling around with the camera trying to get the memory card in and set everything up, I could begin. At first, I wasn’t sure how exactly to take the pictures, even though it is rather straightforward. You just pick a distance from the object and start from a particular angle – I decided to start at the front. It was a little uncomfortable because I had to kneel on the ground, where we had put the Grogu. We decided this because one of the requirements when taking the pictures was to have flat lighting and the one place where we could do this in the time given was the ground. A proper set-up would have consisted of a small, portable table that you could cover with a plain, smooth material and that you could move somewhere with ideal lighting and rotate, but we didn’t have the time or the equipment to build something like that.
It was a lot harder than I expected since I had to gauge how far to move each time and where exactly to position the camera, with almost no memory of the previous photos taken since there were so many. I tried my best to move in small increments but I think that I overestimated how much overlap there would be since I didn’t get 80 photos, which we were told was our minimum. After getting the main turnaround, I had to go again but with a higher angle in order to capture the top as well. That was where things got even messier since the lighting fluctuated a lot.
In fact, looking back at our photos now, I think that was a major issue since some of them are quite under-exposed and dark whether others are brighter. Despite this though, I think that Will’s data set of Tegan will be most successful. He really focused on the overlap and moved in tiny segments each time he took a picture. Consistency is the hardest thing to achieve and human error will almost always be present but I think that he did a very good job overall.
It was a little difficult to consider this a group activity since only one person was needed to take pictures at a time and the rest of us could only really stand around and occasionally point something out. Because of this, I decided to try the technique again on my own, not being satisfied with the first. I found a random object, my pencil case, and went to a different area with decent lighting to capture a new image data set. I used my phone since we didn’t have any more available cameras and tried my best to take more photos than last time at a higher quality. Once again, I found it difficult to have to do so on the floor since I kept having to scoot awkwardly around if I wanted to keep the distance around equal all of the way around.
I also noticed that the longer I spent on the set, the lazier I got towards the end. It became harder and harder to know if I had achieved overlap correctly and my photos became more sporadic and random. I think this is also why we had fewer pictures of the Red Bull than of Tegan. I really like making things look aesthetic so I set the cans around the box in a way that resembles commercials, maximising the presentation of the product. We even sprayed some water droplets on top to make them look more appealing.
We jumped around quite a lot with this one and I can see from the photos in our group’s drive folder that there are some very zoomed out shots. I am curious to see how the model will look, and how the reflective surfaces of the cans affect the reconstruction process. I also think that it would have been better to shut the door. We were filming near the entrance of the classroom and you can see the light spilling in from the corridor and making a highlight on the ground near the set-up. I suppose the pictures can be cropped and altered in the clean-up stage and this will be removed but it would have saved us some time to make sure the environment was more prepped beforehand.
Lastly, there was the shoe. It was Tegan’s turn to take pictures all around the object and we moved to the area outside of the classroom for this. The lighting there was better but some of Tegan’s photos are quite blurry. She also decided to zoom in and get quite close to the shoe when taking the pictures, particularly at the front, which will make for an interesting model! I like that you can see all of the mud in greater detail.
None of our image sets includes the bottom of the object, which I do wish we had tried at least once since it makes for a more realistic 3D scan. At the time, we didn’t add any markers/squiggles either but I am not quite sure what kind of difference that would have made, considering that the objects themselves are quite distinguishable, the ground had a wooden pattern and we also moved around quite a lot in terms of distance from the object. I am looking forward to seeing how all of these objects will look when reconstructed!
To conclude, our lesson was productive albeit a little unpolished and we have a lot of images to work with. We didn’t quite hit the minimum 80 mark for everything in my opinion (I haven’t counted) but with all of us working as a team on cleaning up the photos, I think that we can get some cool photogrammetry models by the end of the process.