Abstract:
Autonomous driving requires accurate and fast point cloud 3D object detection. One-stage 3D object detection methods can achieve real-time performance, but they are dominated by anchor-based detectors that are inefficient and require post-processing. This paper eliminates anchors and models an object as a single point—its bounding box center. An anchor-free CenterNet3D network detects 3D objects using the center point. CenterNet3D directly regresses 3D bounding boxes and finds center points using keypoint estimation. Due to point cloud sparsity, 3D object center points are likely to be in empty space, making boundary estimation difficult. We suggest adding a corner attention module to force the CNN backbone to focus on object boundaries. We also develop an efficient keypoint-sensitive warping operation to align the confidences to the predicted bounding boxes for one-stage detectors. Our non-maximum suppression-free CenterNet3D is more efficient and simpler. CenterNet3D is tested on KITTI and nuScenes. Our method outperforms all state-of-the-art anchor-based one-stage methods and is comparable to two-stage methods. It achieves the best speed-accuracy trade-off at 20 FPS.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
