Enhanced Sparse Convolutional Detection Model for 3D Object Detection in Autonomous Vehicles Adapted to Traffic Conditions in Vietnam

The paper introduces a novel deep-learning model for enhancing 3D object detection in autonomous vehicles tailored to Vietnamese traffic conditions. The model improves efficiency by computing features only for relevant voxels, reducing computational costs and accelerating convergence through direct point cloud transformations. To address the limitations of the KITTI 3D dataset, a new proprietary dataset was created using Lidar16 and Lidar32 sensors, capturing specific Vietnamese traffic conditions. Experimental results on both the KITTI 3D and Phenikaa-X datasets show promising improvements in object detection performance, advancing techniques in 3D object detection and enhancing safety and reliability in diverse traffic environments.