OBJECT CLASSIFICATION USING CNN-BASED FUSION OF VISION AND LIDAR IN AUTONOMOUS VEHICLE ENVIRONMENT

Abstract

This study presents a vision-based, light-detection-based, range-fusion-based object 
classification system to aid autonomous vehicles in their navigation. This method is 
based on the ideas of convolutional neural networks (CNNs) and image upsampling. 
The upsampled LIDAR data is fed into a deep convolutional neural network (CNN) 
together with RGB data to generate point clouds, which are then transformed into 
depth information at the pixel level. This method might be beneficial for autonomous 
vehicle object categorisation by combining vision and LIDAR data to derive feature 
representations. It is also utilised to guarantee minimal loss item classification. The 
results of the experiments reveal how effective and efficient item classification 
systems are.