I am working on a visual lane detection system which is required to predict the direction of the road. Possible outcomes are l, s, and r (left, straight, and right). Say the last 10 outcomes are given by
outcomes = [l, l, l, l, s, s, s, l, l, s];
and the lane detection code generates a new outcome as s. How can I check the accuracy of the new outcome based on my last 10 outcomes by given a greater weight to the latest outcomes?
Note: I am using a threshold value (the difference between the slopes of left and right lanes) which I manually tuned to distinguish between l, s, and r. Although, this approach generally works fine, it also causes a fluctuation between two successive states ( l, s, and r) during transitions. I am trying to eliminate this issue by employing a probabilistic approach.