Work:
- Worked on color detection algorithm for pink tape wrapped objects.
- I fine-tuned the HSV range for neon pink to work better under different lighting and shadow conditions. I explored several ranges using the taskbars I created last time and managed to almost eliminate glare from most tests as well.
- I wanted to solve segmentation/fragmentation this week so I explored several methods that I found online and received from the professor/Omar.
- The first method I attempted to use was morphological contours. I tried to use OpenCv’s erode and dilation functions to try to close/cluster my contours for objects located at further distances from the camera. I was not able to get a noticeable result from running several tests so I think I was not fully doing the procedure correctly. I will try to continue testing this method next week after the demo.
- The next method I tried was downsampling with the contour drawing, but that also did not really produce any noticeable results. I will ask Omar for help with that next week.
- My final method and the one I had results with was applying a Gaussian Blur to the image before attempting to find contours. In my case, I used a 5×5 G. Blur for my webcam (low resolution) and was able to eliminate a lot of the noise and fragmentation that I saw next week. Combined with having a better tested HSV range, I was much more satisfied with my testing as shown below.
- I added taskbars for moving the “bin” rectangle bounding box around.
- I added text representing the number of items within the bin.
- Items within the bin would not have contours or bounding boxes drawn around them but would be “counted” and updated in the text showing # of items in the box.
- I also updated each contour to have a bounding box drawn around it, a centroid dot in the middle, coordinates displayed better on top, and a number indicating the order and number in which it appeared on the video feed.
- Finally, I worked on adding information to our slideshow for the demo next week.
Future:
- Next week, I want to continue fine tuning the HSV range for neon pink in bright lighting. Additionally, I want to continue to run testing on downsampling and morphological contours in addition to any other methods I research for color thresholding. I will also move onto testing actual tools on a dark-tiled floor next week to simulate the real workshop environment. Finally, I will also give the demo with my teammates.
Here are some screenshots from my webcam during live video feed color detection testing.