Aiming at solving the problem of target occlusion in crowded scenes, the aggregation of multiple camera views can outperform a single view. Some researchers project the feature map or RGB image getting from the feature extraction network to the ground plane by perspective transformation and then use the spatial aggregation without anchor frame to aggregate the feature maps of multiple views. However, the projected feature maps are limited by a lack of semantic information. In order to obtain feature maps with rich semantic information, we will introduce a new multi-view detection method MVT, which effectively extracts features from multiple perspectives. To improve TP and the overall robustness of the model, we will also introduce a method for predicting the positioning uncertainty of bounding box reliability.
Vehicle-Road Multi-View Interactive Data Fusion
Multi-sensor Fusion for Autonomous Driving ; Kapitel : 8 ; 177-190
2023-05-11
14 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
INTERACTIVE MULTI-DISPLAY SURROUNDING-VIEW SYSTEM FOR VEHICLE
Europäisches Patentamt | 2023
|Edge-side multi-sensor data fusion system of vehicle-road cooperative system
Europäisches Patentamt | 2021
|Multi-data-source data fusion processing method and device based on vehicle-road cooperation
Europäisches Patentamt | 2021
|Vehicle-road cooperation data fusion system and method
Europäisches Patentamt | 2023
|Unmanned underwater vehicle based on seabed multi-view fusion monitoring
Europäisches Patentamt | 2023
|