Utilized the KITTI dataset to process over 10,000 point cloud images, resulting in a highly accurate obstacle detection system with a precision rate of 96%. Leveraged ResNet and Feature Pyramid Networks (FPN) algorithms to extract robust features from point clouds, improving object classification accuracy by 18% over previous methods. Spearheaded the implementation using ROS and Rviz, streamlining real-time visualization of LIDAR and camera data, and reducing the risk of accidents by up to 40% through faster obstacle identification and localization.
This project enhances vehicle safety by providing real-time obstacle detection and visualization in autonomous driving systems.