Work:
- This week I focused on final tuning/adjustments of the color algorithm and integration with the final robot.
- For the OpenCV interface we had a rescoping of the robot so I worked on making the bounding boxes change color when they are within grabbing range of the arm.
- I also worked with Omar on compiling and integrating the code with the Jetson and the RevHub on the final robot and debugging the process.
- I also worked with my teammates on finishing the final presentation by adding the updated design components (Block diagram, CAD model, and demo videos).
- I also tested my color thresholding subsystem more extensively:
- Ran this over 30 times from different angles and 20 taped-wrapped tools/pencils/pens to obtain metrics (15 times in bright light and 15 times in dim/shadowy light)
- I observed a < 8% false positive rate under uneven or shadowy lighting conditions (red pen was detected 23 over 300 tests)
- < 3 % false negative rates under uneven or shadowy lighting conditions (pink tools not detected only 11 times out of 300 tests)
- Otherwise, < 1% false negative and false positive rate under bright lighting conditions ( 3 red detections and 2 pink false negatives out of 300+ tests)
Future:
- Next week is important as I will be giving the final presentation and working with my teammates on the poster and final demo video.
Here are some screenshots from my webcam during live video feed color detection testing.