I am using same solution, and getting right count with test input, but the actual input gives too many.
Annoying, and really hard to debug. I made renderer to visualize, but unable to find the bug.
One misunderstanding was that I counted same walls twice, because the result should not count same added wall twice if it has same x,y.
You can track by just keeping list of all positions and movement direction, if two are same, it means that it is in a loop.