Enhancing 3D Object Detection in Autonomous Vehicles: Multi-Sensor Fusion with Attention Mechanisms

• Eduardo Sperle Honorato USP
• Denis Fernando Wolf USP

Abstract

In the realm of Autonomous Vehicles (AVs), effective 3D object detection is paramount for ensuring safe navigation in complex environments. The integration of data from multiple sensors, such as cameras and LiDAR, presents challenges in accurately perceiving the surrounding environment. In this paper, we propose several enhancements to the BEVFusion model, a state-of-the-art approach for fusing camera and LiDAR data for 3D object detection in AVs. Specifically, we investigate the integration of attention mechanisms to improve sensor fusion within the BEVFusion framework. Through extensive experiments on the nuScenes and nuScenes mini datasets, the best-performing model from our proposed approaches achieved a relative improvement of 1.2% in mAP and 0.6% in NDS compared to the baseline model. These findings highlight the effectiveness of our attention-based fusion strategy in enhancing detection accuracy, making it a robust solution for real-world autonomous driving scenarios.

Keywords: Measurement, Solid modeling, Attention mechanisms, Three-dimensional displays, Laser radar, Accuracy, Computational modeling, Object detection, Cameras, Autonomous vehicles, Attention Mechanisms, 3D Object Detection, Autonomous Vehicles, Sensor Fusion