LaneNet: Real-Time Lane Detection Networks for Autonomous Driving

  • 2018-07-04 18:05:04
  • Ze Wang, Weiqiang Ren, Qiang Qiu
  • 44

Abstract

Lane detection is to detect lanes on the road and provide the accuratelocation and shape of each lane. It severs as one of the key techniques toenable modern assisted and autonomous driving systems. However, several uniqueproperties of lanes challenge the detection methods. The lack of distinctivefeatures makes lane detection algorithms tend to be confused by other objectswith similar local appearance. Moreover, the inconsistent number of lanes on aroad as well as diverse lane line patterns, e.g. solid, broken, single, double,merging, and splitting lines further hamper the performance. In this paper, wepropose a deep neural network based method, named LaneNet, to break down thelane detection into two stages: lane edge proposal and lane line localization.Stage one uses a lane edge proposal network for pixel-wise lane edgeclassification, and the lane line localization network in stage two thendetects lane lines based on lane edge proposals. Please note that the goal ofour LaneNet is built to detect lane line only, which introduces moredifficulties on suppressing the false detections on the similar lane marks onthe road like arrows and characters. Despite all the difficulties, our lanedetection is shown to be robust to both highway and urban road scenarios methodwithout relying on any assumptions on the lane number or the lane linepatterns. The high running speed and low computational cost endow our LaneNetthe capability of being deployed on vehicle-based systems. Experiments validatethat our LaneNet consistently delivers outstanding performances on real worldtraffic scenarios.

 

Quick Read (beta)

loading the full paper ...