Processing a point cloud generated by an environment detection device of a motor vehicle to a Poincaré-invariant symmetrical input vector for a neural network

摘要

The invention relates to a method for processing a point cloud (4) generated by an environment detection device (3) of a motor vehicle (1), the point cloud (4) representing an environment (5) of the motor vehicle (1) and a set of points (6). in a spatial space (7), with a) providing the point cloud (4) by the environmental detection device (3) to a computing device (2); b) mapping the point cloud (4) into a mathematical feature space (8, 9) in which objects Poincareetransformations-invariant or permutation-invariant are represented by the computing device (2) by applying a predetermined transformation rule (Q, F) on the point cloud (4); and c) providing a feature vector (10) representing the point cloud (4) in the mathematical feature space (8, 9) as an input vector to a neural network (11) for processing the point cloud (4) through the neural network (11).