Camera parameter estimation using semantic labels

In one implementation, a method of determining an estimate of a position of a camera is performed at a device including one or more processors and non-transitory memory. The method includes obtaining a point cloud of a physical environment including a plurality of points, wherein each of the plurality of points is associated with set of three-dimensional coordinates in a three-dimensional coordinate system of the physical environment, wherein the plurality of points includes a first cluster of points associated with a first semantic label. The method includes obtaining a two-dimensional image of the physical environment associated with a point-of-view set of three-dimensional coordinates in the three-dimensional coordinate system of the physical environment. The method includes detecting, in the two-dimensional image of the physical environment, a representation of a first object corresponding to the first semantic label. The method includes determining a first plurality of sets of two-dimensional coordinates in a two-dimensional coordinate system of the two-dimensional image of the physical environment corresponding to the representation of the first object. The method includes determining, from the first cluster of points associated with the first semantic label, a first plurality of sets of three-dimensional coordinates in the three-dimensional coordinate system of the physical environment respectively corresponding to the first plurality of sets of two-dimensional coordinates in the two-dimensional coordinate system of the two-dimensional image of the physical environment. The method includes determining an estimate of the point-of-view set of three-dimensional coordinates based on the first plurality of two-dimensional coordinates and the first plurality of three-dimensional coordinates.