Method and device for constructing 3D scene model

A system for constructing a 3D scene model, comprising an acquisition device and a server, wherein

• the acquisition device is configured for acquiring images of a target scene and sending an acquired current frame to the server;
• and
• the server is configured for receiving the current frame and converting the current frame into a first point cloud;
• determining a first partial point cloud corresponding to a dynamic object from the first point cloud;
• constructing a foreground 3D model based on the first partial point cloud;
• and superimposing the foreground 3D model with a pre-obtained background 3D model to obtain a current 3D scene model, wherein the background 3D model is constructed based on static objects in the target scene,
• wherein the acquisition device is further configured for scanning the static objects in the target scene to obtain scanned data, and sending the scanned data to the server;
• and
• the server is further configured for:
• receiving the scanned data, converting the scanned data into a second point cloud, determining a reference frame point cloud and each of offset frame point clouds in the second point cloud;
• determining an offset coefficient of each offset frame point cloud relative to the reference frame point cloud;
• for each offset frame point cloud, fusing the offset frame point cloud to the reference frame point cloud based on the offset coefficient corresponding to this offset frame point cloud, to obtain a fused point cloud;
• and constructing the background 3D model based on the fused point cloud;
• wherein fusing the offset frame point cloud to the reference frame point cloud comprises:
• transforming coordinates of the offset frame point cloud based on the offset coefficient corresponding to the offset frame point cloud;
• determining whether the transformed offset frame point cloud and the reference frame point cloud satisfy a closed loop principle;
• if the transformed offset frame point cloud and the reference frame point cloud do not satisfy the closed loop principle, fusing the transformed offset frame point cloud and the reference frame point cloud;
• and
• if the transformed offset frame point cloud and the reference frame point cloud satisfy the closed loop principle, correcting the reference frame point cloud by using a preset algorithm and fusing the transformed offset frame point cloud and the corrected reference frame point cloud.