Coupled Object Detection and Sparse Depth Estimation

摘要

Detecting and localizing other traffic participants, such as pedestrians and vehicles, from a moving vehicle has various applications in smart vehicle. In this work, we address such a problem by utilizing image sensors, namely stereo cameras mounted on the vehicle. Our proposed method integrates appearance based object detection and sparse depth estimation in a novel fashion. With depth estimation, we transform the prior distribution of objects' actual size into the distribution of their imaged size to improve the detection performance. By contrast, we use depth information that contributed to correct object hypotheses to better localize objects. Being different with many previous works, we take the trade-off between accuracy and computational cost in the first place of consideration, and try to make the most efficient integration for onboard applications.