3D target detection algorithm based on central correction network and decoupling detector
A 3D target detection algorithm based on central correction network and decoupling detecting probe is proposed.Firstly,three-channel image block representation is used to process the three-dimensional(3D)input information,and two-dimensional(2D)detection frame is used to filter the background points,so as to reduce the amount of data.Then,the center correction network based on CSP Darknet module is used to align the center of the obtained image block to the real center in the target coordinate system.Finally,a decoupling detecting probe based on 1×1 convolution is used to classify and regress the categories and residual values of the parameterized 3D bounding boxes.The experimental results show that the 3D average precision of the improved detection algorithm in the simple,medium and difficult modes of KITTI dataset is improved by 2.82,5.54 and 3.81 than that of the same type of algorithm,respectively,and better detection results are obtained.
NETL
NSTL
万方数据
