Enhanced Point Cloud Object Classification using Convolutional Neural Networks and Bearing Angle Image

Convolutional neural networks, a type of deep neural network, have recently achieved great success across various fields, including image recognition and natural language processing. Their unique structure, designed to process grid-like data, makes CNNs particularly effective for tasks that require pattern recognition and spatial understanding. In order to identify items in a point cloud that was acquired by LiDAR on city streets, a unique framework using CNN is proposed in this research. Discrete point cloud objects are transformed into pictures with bearing angles (BAs). CNN is then used to classify things based on BA images. This paper's primary contribution is the efficient recognition approach that just makes use of point cloud data. However, as most 3D object classifiers combine color pictures and point clouds, their models require enormous amounts of memory to store the parameters and are quite complex. The proposed technique improves both time efficiency and memory usage by detecting ground points and clustering objects, processing all points in scanline-major and layer-major order. High classification accuracy of 96.0% is attained by the suggested BA-CNN classifier.