The application relates to the technical field of industrial
automation, and discloses a car compartment environment real-time modeling method based on three-dimensional vision fusion, which comprises the following steps: obtaining three-dimensional
perception data of the inside of a car compartment, associating observation
direction information with each
perception point, and voxelizing and dividing the space; on this basis, the observation direction consistency in the
voxel unit and the space
accessibility based on the car compartment entrance are analyzed respectively; the
voxel unit is determined in terms of
real structure and risk missing structure by combining the consistency parameter and the
accessibility parameter, and the risk missing structure is compensated in combination with the spatial continuity of adjacent stable structures; meanwhile, the
voxel occupation state is compared for consistency between adjacent scanning periods, structure stability information is generated, and the three-dimensional model of the car compartment is corrected in real time according to the structure stability information, so that a car compartment environment model suitable for loading planning is dynamically output during the loading operation.