Stixel optimization: Representing challenging on-road scenes

This paper presents a stereo-based method for robust vertical road profile detection. The aim is to ensure a very low false-detection rate for obstacles in challenging datasets. Basically, frames are segmented into ground manifold and obstacles. To robustly estimate a vertical road profile, our method applies cuts through a v-disparity matrix along columns to achieve a minimal cost specified by the matrix itself while maintaining a desired smoothness; the minimization is based on the Viterbi algorithm, a dynamic programming technique. Experiments illustrate the performance of the proposed method using available datasets. Results show that the proposed method outperforms two previously published methods on the used datasets. In particular, its performance is superior to the others in cases of challenging weather conditions.