This study presents an asymmetric late fusion approach to fuse camera and lidar outputs from different neural networks. By fusing two different modalities, each network's false positive results are eliminated and a superior vector space compared to single modality methods is created. Proposed 3D bounding boxes from lidar networks are supported with 2D segmentation networks that work with a camera. Ground plane provided by the lidar input is also detected to eliminate possible 2D camera drivable area false positive results, namely, roadblocks, grass, and side of tunnel walls. 3 camera networks and 6 different 3D lidar networks tested with KITTI dataset illustrate that the presented fusion approach exceeds the performance of Lidar-only methods up to 9.87% category wise. Comparison with two different fusion approaches also show our superior performance.
A Late Asymmetric Fusion Approach To Eliminate False Positives
2022-10-08
767504 byte
Conference paper
Electronic Resource
English
Use of Tinted Reflectors to Eliminate False Positives in Adaptive Lighting Control Systems
SAE Technical Papers | 2009
|Systems and methods for detecting false positives in collision notifications
European Patent Office | 2025
|ELIMINATION OF FALSE POSITIVES IN VIV IDENTIFICATION USING MACHINE LEARNING
British Library Conference Proceedings | 2019
|SYSTEMS AND METHODS FOR DETECTING FALSE POSITIVES IN COLLISION NOTIFICATIONS
European Patent Office | 2024
|