Dynamic 3D Scene Analysis from a Moving Vehicle

In this paper, we present a system that integrates fully automatic scene geometry estimation, 2D object detection, 3D localization, trajectory estimation, and tracking for dy- namic scene interpretation from a moving vehicle. Our sole input are two video streams from a calibrated stereo rig on top of a car. From these streams, we estimate Structure- from-Motion (SfM) and scene geometry in real-time. In par- allel, we perform multi-view/multi-category object recogni- tion to detect cars and pedestrians in both camera images. Using the SfM self-localization, 2D object detections are converted to 3D observations, which are accumulated in a world coordinate frame. A subsequent tracking module an- alyzes the resulting 3D observations to find physically plau- sible spacetime trajectories. Finally, a global optimization criterion takes object-objectinteractions into accountto ar- rive at accurate 3D localization and trajectory estimates for both cars and pedestrians. We demonstrate the perfor- mance of our integrated system on challenging real-world data showing car passages through crowded city areas.