0.3 4:conclusions and future steps

4.0 conclusions

As far as the Hough Transform goes, our findings were as follows:

• The success of the algorithm depends on edge magnitude threshold, Hough transform threshold, and radius parameters. and image noise. Changing thresholds results in tradeoffs between false positives and false negatives.
• For images with single balls, the algorithm can detect the ball with 95.6 % accuracy. Of the wrongly detected images, 66.7% had high amounts of motion blur.
• The algorithm can also detect balls partially cut out of the image as long as more than 60% of the ball is within the image.
• In images where multiple balls are close or overlapping, the algorithm may mistake the two balls as a single object or miss one of the balls.

The video linked in the Additional Resources section also demonstrates our Hough transform implementation in action.

4.1 future steps

The next primary addition to our robot is object permanence, giving it the capability to remember balls that have moved out of its field of view and rotate itself to find them again. Other future goals include recognizing when an obstacle is in front of the ball, fine-tuning the PID control in order to speed up the angling and approach to a ball, and being able to prioritize different balls based on properties like color.

4.2 additional resources

A video of our robot in action can be found here on Youtube.

Our poster presentation can be downloaded from this link .

Python code for the Hough transform can be found here .

The custom library used by the robot can be found here .

4.3 references

• M.K. Vairalkar and S.U. Nimbhorkar. Edge detection of images using Sobel operator. International Journal of Emerging Technology and Advanced Engineering, 2(1):291–293,2012.
• R.O. Duda and P.E. Hart. Use of the Hough transform to detect lines and curves in pictures. Community Assoc. comput. Mach. 15 (1975), pp. 11-15.