Skip to content
Conference

A multimodal 3D object detection framework based on hybrid radar point generation and dual synchronization fusion

Nov 2026 · International Conference on Algorithms, Image Processing, and Deep Learning (AIPDL 2026) · 0 citations

Abstract

Aiming at the instability of 3D object detection caused by sparse radar point clouds and illumination-susceptible image features in complex traffic scenarios, this paper constructs an intelligent perception algorithm system based on radar-vision collaboration. Taking hybrid radar point generation, distributed point feature encoding and dual-synchronized BEV fusion as its core computational pipeline, the system operates as follows. First, image semantics are utilized to constrain radar foreground regions, and enhanced point clouds are generated via Gaussian-uniform hybrid sampling. Subsequently, feature representations of raw points, foreground points and generated points are distinguished. Finally, cross-modal feature alignment and dynamic weighted fusion are realized through spatial synchronization and modal synchronization. Experimental results show that the proposed method achieves an EAA AP of 58.96 and an RoI AP of 79.46 on the VoD validation set, as well as a 3D mAP of 37.21% and a BEV mAP of 43.23% on the TJ4DRadSet test set, outperforming multiple mainstream radar-vision fusion models. The results demonstrate that the proposed algorithm system can improve the complementary utilization efficiency of multisource perception information and maintain favorable detection robustness and application universality under complex environments such as low light and light reflection.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.