基于Transformer的道路场景点云分类与分割方法

针对多目标识别过程中点云分类和分割精度不高的问题,提出了一种基于改进Transformer模型的点云分类与分割方法DRPT(Double randomness Point Transformer),该方法在Transformer模型卷积投影层创建新的点嵌入,利用局部邻域的动态处理在数据特征向量中持续增加全局特征属性,从而提高多目标识别中点云分类和分割的精度。实验中采用了标准基准数据集(ModelNet40、ShapeNet部分分割和SemanticKITTI场景语义分割数据集)以验证模型的性能,实验结果表明:DRPT模型的pIoU值为859,比其他模型平均高出35,有效提高了多目标识别检测时点云分类与分割精度,是对智能网联技术发展的有效支撑。

A Transformer based classification and segmentation approach for classifying and segmenting road field attraction clouds

To address the problem of low accuracy of point cloud classification and segmentation in the process of multi target recognition,a point cloud classification and segmentation method DRPT(Double randomness Point Transformer)based on the improved Transformer model is proposed in this paper.The approach creates new point embeddings in the convolutional projection layer of the Transformer model and uses local dynamic processing of local neighborhoods to continuously add global feature attributes in the data feature vector,thus improving the accuracy of point cloud classification and segmentation in multi target recognition.Standard benchmark datasets(ModelNet40,ShapeNet partial segmentation and SemanticKITTI scene semantic segmentation datasets)are used in the experiments to validate the performance of the model.The experimental results show that the pIoU value of the DRPT model is 85.9%,which is 3.5% higher than other models on average,and effectively improves the accuracy of point cloud classification and segmentation during multi target recognition detection,which is an effective support for the development of intelligent network technology.