Work:
- Worked on color detection algorithm for specifically pink-colored tape
- I managed to find the HSV range that could detect neon pink. As shown in the video below, I was able to successfully differentiate mostly between the neon pink tape, my lip color, and the light pink of the fork. There are a few pixels that are sometimes detected elsewhere that do not match “neon pink” that I will continue to calibrate.
- Segmentation of the contours was a big issue, so I aimed to solve that this week. One solution for now is to only draw contours around areas that are of at least a certain size (100 px) for now. This helps reduce segmentation where the tape or object is fragmented into multiple areas with multiple contours. This also helps avoid drawing “mini contours” around very small pixels areas that sometimes are registered but not the tape-wrapped object.
- Lastly, I figured out how to use:
- Trackbars to control the lower and upper HSV ranges. Before I was very inefficiently manually changing the HSV values and recompiling each time to test the masking.
- Getting and drawing the coordinates of the centroid of each contour on the live camera/frame feed.
- I also managed to set up my Kinect sensor/camera. I ran into a lot of issues setting it up on Windows with archived packages and outdated drivers and it would not connect to OpenCV at all or even output an image.
- I dug around online and eventually setup up a Ubuntu VM to connect to my Kinect. After a lot of trial and error I managed to connect the Kinect successfully to my VM, get a working image output, connect to OpenCV, and see the RGB image and the depth image.
Future:
- Next week, I want to continue fine tuning the HSV range for neon pink in bright lighting. Additionally, I want to nearly completely eliminate segmentation and any false positives while drawing bounding boxes around each contour. I also want to be able to detect neon pink objects from far away (more preciseness needed) and integrate the depth sensor of the Kinect camera as well. I also want to remove an area on the feed for contour/color detection as that will be our “basket” for already collected items. Finally, I will help Omar/Hojun with the software/hardware integration as we prep for our Demo.
Here are some screenshots from my webcam during live video feed color detection testing.
