📄 中文摘要
随着汽车配备越来越多的摄像头，单张照片无法全面捕捉周围环境。新方法BEVDet4D使摄像头系统能够将当前场景与稍早前的场景进行比较，从而更好地捕捉运动和方向。通过简单的步骤融合旧视图和新视图，该系统能够读取之前隐藏的时间线索，且成本几乎没有增加。这大幅提高了速度预测的准确性，模型显著降低了速度误差，使得仅依靠视觉的系统能够接近使用高级传感器的系统。该方法支持多镜头同时工作，使得多摄像头汽车能够更平滑地跟踪车辆和行人。在一次大型测试中，该系统表现出色。
BEVDet4D：利用时间线索进行多摄像头3D物体检测
出处: BEVDet4D: Exploit Temporal Cues in Multi-camera 3D Object Detection
发布: 2026年2月11日
📄 English Summary
BEVDet4D: Exploit Temporal Cues in Multi-camera 3D Object Detection
With the increasing number of cameras in vehicles, a single photo fails to capture the surrounding environment comprehensively. The new approach, BEVDet4D, enables camera systems to compare the current scene with one from a moment ago, allowing for better detection of movement and direction. By adding a simple step to fuse the old and new views, the system can read previously hidden temporal cues with minimal cost increase. This significantly enhances the accuracy of speed predictions, as the model greatly reduces velocity errors, bringing vision-only systems closer to those using advanced sensors. The method works with multiple lenses simultaneously, enabling multi-camera vehicles to track cars and pedestrians more smoothly. In a large-scale test, the system performed exceptionally well.