The invention belongs to the technical field of
computer vision, and particularly relates to a
street scene ground object real-time semantic segmentation and geographic positioning method and
system based on a camera, and the method comprises the steps: S1, carrying out the image collection and preprocessing; s2, performing semantic segmentation; s3, panoramic
monocular depth
estimation and scale
recovery are carried out; s4, target detection and pixel extraction are carried out; step S5, resolving a space coordinate; and S6, performing Kalman filtering fusion based on a
time sequence, and outputting a final convergence coordinate of a filter as a final geographic position of the ground feature. Different from a
vehicle positioning technology which depends on a prior high-precision map for matching, the method provided by the invention has the advantages that a panoramic depth
estimation and space coordinate resolving
algorithm is utilized, the three-dimensional space position of a ground object can be directly deduced reversely from two-dimensional image pixels under the condition of no prior map data, and the WGS84 absolute geographic coordinate of the ground object is calculated in real time in combination with GNSS / IMU data.