See how multiple floor cleaners share environment images to build topological maps, treat proce
Sparse voxel attention lifts 2D camera images into 3D semantic voxels, reducing ambiguity and 3D CNN complexity for scene completion.
Forward-backward view transformation improves 3D voxel occupancy prediction accuracy and efficiency for autonomous vehicle motion planning.
Combining surround-view video with ultrasonic sensing highlights open parking slots and supports selectable autonomous parking with flexible viewpoint changes.
Bird's-eye image annotation with extended 3D areas cuts training data needs for robust obstacle detection without complex sensor hardware.
Driving-status-based crop enlargement widens the rearward view during turns while avoiding a fixed loss of image detail on the in-vehicle display.
Selective sensor sub-areas let one stereo fisheye camera switch field of view during operation, balancing depth accuracy, coverage, and computing load.