Work:
- Worked on tuning color detection algorithm for pink tape wrapped objects.
- I fine-tuned the HSV range further for neon pink to work better under different lighting and shadow conditions. I also adjusted the increased blurring by (3,3) and the value min to eliminate detection of red colors (mostly).
- With the thresholding completed, I focused on testing my rudimentary system. I put my camera on top of a box on top of a cleaning Roomba and simulated moving around, detecting objects, and moving towards said objects.
- The camera & algorithm was able to successfully detect pink-taped wrapped “tools” on the well lit floor and draw bounding boxes around each one and label them in order of time seen (distance essentially).
- For the actual distance, I was not able to “detect” distance, as I do not have a depth camera. It should only take 1-2 lines of code to add the distance metric when used on the Intel Depth Camera later during integration.
- I also conducted a basic simulation of having a bin at the front of the Roomba where items are already present and deposited during collection. The live feed was able to still successfully count and draw around the objects on the floor while simultaneously counting the number of collected tools already in the bin.
- Overall, the testing was successful and the algorithm is mostly completed with minor tweaks needed for different lighting or color thresholds as needed.
Future:
- Next week is important as I will work with my teammates to complete integration of all our finished sub-systems to produce a working robot that can move towards detected tools and pick them up.
Here are some screenshots from my webcam during live video feed color detection testing on the “moving” Roomba.