基于多台飞行时间相机的建图与结构语义的三维库位检测

Threedimensional Parking Slot Detection Using Mapping and Structural Semantics Based on Multiple Timeofflight Cameras

摘要:

提出一种基于多台飞行时间(ToF) 相机的建图与结构语义的三维库位检测方法.利用多台ToF相机进行联合观测，采用视觉里程计对局部停车场场景进行建图，并通过对停车场墙面、顶面、地面和障碍物结构语义信息的聚类和分割，实现对三维停车库位的实时检测.结果表明：该方法在有效探测范围内库位检出率为94.83%，库位宽度识别精度为14.4 cm，高度识别精度为12.4 cm.

Abstract:

A detection method for threedimensional parking slot was proposed by using multiple timeofflight (ToF) cameras. The local scene of a parking lot was jointly captured and mapped by using multiple ToF observations in real time, and structural semantics such as walls, ceilings, floors and obstacles were segmented in the scene for the detection of parking slots. The detection rate of 94.83% was achieved in the experiment. The precision of detected width and height of parking slots are 14.4 cm and 12.4 cm respectively.