Multi lane detection, curve fitting and lane type classification

Abstract

An advance driver assistance system has been proven to be effective in reducing vehicle accidents. For autonomous driving cars continuous navigation of the road is the most desired functionality. Such systems help to avoid road accidents and increase driver safety by alerting the driver prior to road scenarios. A key feature in self or autonomous driving vehicles is detecting the lane and lane type on the road. Where lane detection is the process of detecting the lanes and providing information on lanes (lane location, lane type and shape). The job of this system is to detect the lane and point out that to the driver so that the driver will be aware of upcoming complicated scenarios such as lane changes. The main objective of this work is to address the challenges faced by lane detection and minimize them. (Such as complex lighting, no lane, shadow occlusion). Our work will help to detect various lanes and distinguish them as continuous, double-dashed and dashed lane types. In this work, a deep neural network called Lane-net is used for lane detection. Image segmentation processes included in this work is Instance segmentation. Segmented data points obtained from lane detection are annotated and processed by a curve fitting mechanism. Lane detection is performed on the TUsimple dataset that supports day, night, and traffic environments. Further, the work continues distinguishing double solid, dashed solid and zigzag lane types. The lanes detected from image segmentation are further classified into continuous, dashed and double dashed lane types.