Method and apparatus for detecting parking area using semantic segmentation in automatic parking system

摘要

To provide a method for learning an automatic parking device of a vehicle for detecting an available parking area.SOLUTION: A method includes the steps of: inputting, if a parking lot image of an area nearby the vehicle is acquired, the parking lot image to output a convolution feature map, outputting a deconvolution feature map by deconvoluting the convolution feature map, and outputting segmentation information by masking the deconvolution feature map; inputting the deconvolution feature map to generate relative coordinates of vertices of a specific available parking region, and generate regression location information by regressing the relative coordinates; and instructing a first loss layer to calculate first losses by referring to the regression location information (ROI) and an ROI GT corresponding thereto, and learning the regressor via backpropagation using the first losses.SELECTED DRAWING: Figure 2