After meeting with Mikki last week, she shared the shark videos she took with her GoPro and lasers setup. She also shared the process she uses to back out the shark dimensions. There's some ImageJ stuff, some video editing, and so on. Her current videos use linear mode, which (I assume) does a pretty good job rectifying the images for processing so that the pixel vectors are more or less parallel. She has some old videos, though, that need calibration.
Anyhow, I figured I might be able to help out with my experience in image processing and analysis. I worked most of the day on making a website that lets you select a video, set dimensions, and measure. I find that I can measure sharks within about 1 cm precision and do it in a few seconds. I haven't shared it with her yet as our next meeting is next week. I'll take some time to polish the interface and add some features. Hopefully, we can get it hosted somewhere so that other people using her shark imaging system can use it to accelerate the database creation effort.
Feels good to find a small team to work with.
Image below shows what the result looks like right now. Not shown are the various buttons and outputs, which need the most cleaning up. The two red dots mark where her parallel laser beam spots fall on the shark. They are calibrated to be 50 cm apart. You can use that information to calculate the centimeters per pixel and base any other pixel-space measurements on it. Of course, it's only approximate because of so many factors. In theory, the biases from things like shark curvature should be somewhat similar and systematic...
Anyhow, I figured I might be able to help out with my experience in image processing and analysis. I worked most of the day on making a website that lets you select a video, set dimensions, and measure. I find that I can measure sharks within about 1 cm precision and do it in a few seconds. I haven't shared it with her yet as our next meeting is next week. I'll take some time to polish the interface and add some features. Hopefully, we can get it hosted somewhere so that other people using her shark imaging system can use it to accelerate the database creation effort.
Feels good to find a small team to work with.
Image below shows what the result looks like right now. Not shown are the various buttons and outputs, which need the most cleaning up. The two red dots mark where her parallel laser beam spots fall on the shark. They are calibrated to be 50 cm apart. You can use that information to calculate the centimeters per pixel and base any other pixel-space measurements on it. Of course, it's only approximate because of so many factors. In theory, the biases from things like shark curvature should be somewhat similar and systematic...
Comments