Road Segmentation and Environment Labeling for Autonomous Vehicles

In autonomous vehicles (AVs), LiDAR point cloud data are an important source to identify various obstacles present in the environment. The labeling techniques that currently available based on pixel-wise segmentation and bounding boxes detect each object road. However, Avs’ decision motion control trajectory path planning depends interaction among objects ability of Avs understand moving non-moving is key scene understanding. This paper presents a novel method combine objects. technique named relational labeling. Autoencoders used reduce dimensionality data. A K-means model provides pseudo labels by clustering latent space. Each label then converted into unary binary labels. These supervised learning methods for segmenting backpropagation network (BPN), along with traditional gradient descent-based methods, Our study evaluated accuracy two as well three layers BPN. two-layer BPN was found be better than three-layer model. According experiments, our showed competitive 75% compared weakly similar area study, i.e., S3DIS (Area 5) 48.0%.