Stereo Tests

Since my last post, the team has run several tests in the Poly Canyon Pool. We have tested all of the ROVs systems and software, including the sonar, the new HOBO data logger,  the ROS logging system, and others. During two of the dives we also attached the GoPros to capture more stereo image for disparity tests.

After our tests we have made the final decision to use MatLab's built in disparity functionally to build our disparity maps. However, our images still require some pre-processing before they can be run though the disparity function. The first step is hand-selecting the best images. Then I crop, resize, color match, and align the images.

Right Image

Left Image

Left Image After Color Match

 Next they are run through MatLab and exported as a gray-scale image. These images are then cropped once again, and finally ready to be used to create geometry.





Below are some sample images of our results.

GoPro Image

Final Disparity Map

As you can see, there is still a significant amount of bad matching. This bad matching is the result of un-unique areas in the images. You can see in the image below that all of the circled areas have similar features and colors.

We are less concerned about this happening in cisterns because of the type of structures we will be focusing on.

Our biggest concern for capturing data right now is lighting. When the ROV does not produce enough light, the stereo images are darker and have less contrast. With less contrast most of the disparity maps turn out noisy and unusable. We have purchase a 95 lumen dive light this year to help shed some light on our subject. I will also be bringing some diffusion filters to spread the light evenly over the surface. Unfortunately, we like likely not have time to test the dive light before we leave. I guess it will be another opportunity to learn by doing!

Check back soon for more updates!