利用多任务级联网络解决6D Pose预测问题

Application of Multi-task Cascade Neural Network in 6D Pose Prediction Problem

针对基于单张RGB(red-green-blue)图像预测目标6D Pose的问题,设计了多任务级联结构的卷积神经网络(convolutional neural networks,CNN)和BBE(bounding box equation)算法实现快速高效的6D Pose预测。在LINEMOD数据集上进行实验,并与LINE2D和Brachmann预测算法进行比较,结果表明,该方法速度和精度均超过LINE2D算法,精度上接近Brachmann算法,但速度更快。

Aiming at the problem of predicting 6 D Pose based on single RGB image,a multi-task cascade convolutional neural network and Bounding Box Equation algorithm are designed to realize fast and efficient 6 D Pose prediction.LINEMOD dataset is used and LINE2 D and Brachmann prediction algorithm are compared-in the experiment.The result shows that the speed and accuracy of the proposed method exceed the LINE2 D algorithm,and the accuracy is close to the Brachmann algorithm,but the speed is faster.