Recent Advancements on Drivable Free Space Estimation Using Monocular Vision

Abstract

This paper presents a overview on recent work on drivable free space estimation with emphasis on monocular vision. Yao et. al. proposed an inference in MRF using various cues based on appearance, edges, spatial and temporal smoothness. Wolcott et. al. added more cues based on perceived motion using optical flow. While Levi et. al. proposed a new column wise regression approach using convolutional neural networks and stixels. All the techniques reviewed in this paper have large processing time, thus seriously limiting their practical application.