An Efficient CNN Architecture for Multilane Detection in Real-Time Using Segmentation

Abstract

Lane detection is the core of an autonomous driver assistance system (ADAS). Detecting lanes makes use of image processing, complex image analysis using Convolution Neural Network, and reducing omission of weak objects by using no pooling operation. CNN is the core for lane detection which helps in identifying complex patterns. The human brain functions in advanced manner so CNN has been designed based on biological neural networks in human brains for mimicking the complex functions performed by human brains. The insensitive identification towards weak objects has been causing a lot of trouble. With only Image detection overcoming this issue was not possible, to overcome this drawback Semantic Segmentation is used which provides precise identification of weak objects. This CNN based architecture for lane detection is not only used to accurately identify visible lane marker but also accurately identify the lanes in challenging scenarios like occlusions, low light, and degraded lane markings. This model is expected to overcome the drawback in previous lane detecting systems.