Method and device for predicting object data concerning an object

A device for determining object data with respect to an object in surroundings of at least one image camera, comprising:

• a processor operatively configured to:
• determine a camera-based feature tensor based on at least one image of the image camera for a first point in time via a neural encoder network;
• transform and/or project the camera-based feature tensor from an image plane of the image to a grid plane of a surroundings grid of the surroundings of the image camera to determine a transformed feature tensor;
• and
• determine object data with respect to the object in the surroundings of the image camera based on the transformed feature tensor via a neural evaluation network,
• wherein the object data comprise one or more predicted properties of the object at a point in time subsequent to the first point in time;
• wherein the processor is further configured to:
• for a plurality of chronologically successive images of the image camera, determine a corresponding plurality of camera-based feature tensors via the neural encoder network;
• wherein the plurality of chronologically successive images covers an acquisition period of time which extends chronologically before and/or up to the first point in time;
• on the basis of the plurality of camera-based feature tensors, determine a corresponding plurality of transformed feature tensors;
• determine odometry data with respect to a movement of the image camera during the acquisition period of time;
• combine the plurality of transformed feature tensors in consideration of the odometry data, in order to determine a combined transformed feature tensor;
• and
• determine the object data with respect to the object in the surroundings of the image camera on the basis of the combined transformed feature tensor via the neural evaluation network.

A method for determining object data with respect to an object in surroundings of at least one image camera, the method comprising:

• determining, via a neural encoder network, a camera-based feature tensor based on at least one image of the image camera for a first point in time;
• transforming and/or projecting the camera-based feature tensor from an image plane of the image to a grid plane of a surroundings grid of the surroundings of the image camera to determine a transformed feature tensor;
• and
• determining, via a neural evaluation network, object data with respect to the object in the surroundings of the image camera on the basis of the transformed feature tensor;
• wherein the object data comprise one or more predicted properties of the object at a point in time subsequent to the first point in time,
• wherein determining the camera-based feature tenor comprises:
• combining a plurality of chronologically successive images of the at least one image camera to form an overall image for the first point in time, wherein the plurality of chronologically successive images were acquired before or at latest at the first point in time by the image camera;
• and
• determining the camera-based feature tensor based on the overall image via the neural encoder network.