Semantic map lightweight method

• Main Authors: QIU SHENG, DONG ZHICHENG, CAI FENGLIN, ZHU XIJIE, LI YING, ZHU QITAO, TOSHIFUMI
• Format: Patent
• Language: Chinese; English
• Published: 07.11.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; PHYSICS

The invention provides a semantic map lightweight method. The method comprises the following steps: extracting a significant trajectory of a region of interest; calculating local feature descriptors along the trajectory, extracting visual features, and performing vector coding on the local feature descriptors; and finally, the encoded features are arranged in a tensor form, and discrimination features are obtained through tucker-3 tensor decomposition, so that the semantic map model is lightened. The method solves the problems that a semantic map is large in data size and large in storage difficulty, the transmission bandwidth, computing power and vehicle-mounted storage of an automobile are limited, and the semantic map and a reasoning model are difficult to operate. 本发明提供了一种语义地图轻量化方法，该方法包括感兴趣区域显著轨迹的提取；沿轨迹计算局部特征描述符，提取视觉特征，并将局部特征描述符进行矢量编码；最后，将编码后的特征以张量形式进行排列，通过tucker-3张量分解得到判别特征，以此实现语义地图模型轻量化。本发明的方法解决了语义地图数据量较大，存在存储难度大，汽车的传输带宽、算力和车载存储受到限制，语义地图和推理模型存在难以运行的问题。