三维目标检测中的先验方向角估计

Prior direction angle estimation in 3D object detection

为了解决基于2. 5维信息进行三维目标检测的方法中方向角的先验信息未得到充分利用的问题,从RGB-D相机获取的深度信息出发,提出一种新的基于二维图像估计先验方向角的三维目标检测方法。通过颜色信息和深度信息得到二维的分割实例,并在分割实例上提取关键点;通过关键点的优化过程排除不确定点和修正误判点;通过点云重建得到关键点的三维坐标,根据关键点的坐标估计目标的方向角,并将其作为初始化三维框的方向角。实验结果表明,所提方法在Amodal3Det方法的基础上提高了1. 3%的平均检测精度。

In order to solve the problem that the prior information of direction angle is not fully utilized in the method of 3D object detection based on 2.5D information,start with the depth information acquired by RGB-D camera,propose a novel 3D object detection method that estimates the prior direction angle based on 2D image.Firstly,a 2D segmentation instance is got through color information and depth information,and the key points are extracted from segmentation instance.Then the uncertain points are excluded and the false points are corrected according to the optimization process of the key points.Finally,the 3D coordinate of the key points are obtained through point cloud reconstruction,and the direction angle of the target is estimated according to the coordinate of the key points,which is used as the direction angle of initial 3D frame.The experimental results show that the proposed method improves the average detection precision of 1.3%on the basis of the Amodal3Det method.