Pixel-level rail transit scene understanding method and system

• Main Authors: GAO YANG, YU HANG, LIU QINGHONG, BAI JIE, GE XUANYU, QIN YONG, LI WEI, CAO ZHIWEI, KOU LINLIN, GUAN JIRUI, WANG YAGUAN, LIAN LIRONG
• Format: Patent
• Language: Chinese; English
• Published: 28.05.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention provides a pixel-level rail transit scene understanding method and system, and belongs to the technical field of rail transit safety operation control, and the method comprises the steps: carrying out the continuous down-sampling operation of an input image through employing a light-weight double-branch fusion backbone network, and extracting image features; and inputting a feature map extracted by the backbone network into a bidirectional feature pyramid pooling network to obtain a multi-scale receptive field and deep semantic features, and further optimizing the scene understanding effect of the method from pixel and edge angles by using a multi-task multi-scale loss function. The lightweight double-branch fusion backbone network can extract image features more quickly and more effectively, the bidirectional feature pyramid pooling network extracts semantic features from a multi-scale angle, a multi-task multi-scale loss function supervises the learning process of the method from two angles of