Learning method and learning device, and testing method and testing device for detecting parking spaces by using point regression results and relationship between points to thereby provide an auto-parking system

A testing method for detecting one or more unoccupied parking spaces for testing by using one or more probability distributions for testing on one or more decision points for testing of the unoccupied parking spaces for testing and relational linear segment information on relationships among the decision points for testing, comprising steps of:

• (a) on condition that (1) a learning device, if at least one parking circumstance image for training, including information on one or more unoccupied parking spaces for training, is acquired through at least one camera for training, has performed (i) a process of instructing a first Convolutional Neural Network (CNN) to apply at least one first CNN regression operation to the parking circumstance image for training, to thereby calculate each of one or more estimated probability distributions for training on each of decision points for training, and (ii) a process of instructing a second CNN to apply at least one second CNN regression operation to the parking circumstance image for training, to thereby generate estimated relational linear segment information for training on relationships among at least part of the decision points for training;
• and (2) the learning device has instructed a loss layer to perform (i) a process of generating a first loss by referring to the estimated probability distributions for training and its corresponding Ground-Truth (GT) probability distributions, and training at least part of parameters in the first CNN by performing backpropagation using the first loss, and (ii) a process of generating a second loss by referring to the estimated relational linear segment information for training and its corresponding GT relational linear segment information, and training at least part of parameters in the second CNN by performing backpropagation using the second loss, a testing device performing (i) a process of instructing the first CNN to apply the first CNN regression operation to a parking circumstance image for testing, to thereby calculate each of one or more estimated probability distributions for testing on each of the decision points for testing, and (ii) a process of instructing the second CNN to apply the second CNN regression operation to the parking circumstance image for testing, to thereby generate estimated relational linear segment information for testing on relationships among at least part of the decision points for testing;
• (b) the testing device performing (i) a process of instructing a first selection layer to generate estimated decision point location information for testing on at least part of the decision points for testing, by referring to the estimated probability distributions for testing, and (ii) a process of instructing a second selection layer to generate estimated linear segment location information for testing on at least part of pairs for testing of at least part of the decision points for testing by referring to the estimated relational linear segment information for testing;
• and
• (c) the testing device instructing a pairing layer to detect the unoccupied parking space for testing by referring to the estimated decision point location information for testing and the estimated linear segment location information for testing.

A learning device for detecting one or more unoccupied parking spaces by using one or more probability distributions on one or more decision points of the unoccupied parking spaces and relational linear segment information on relationships among the decision points, comprising:

• at least one memory that stores instructions;
• and
• at least one processor configured to execute the instructions to perform processes of:
• (I) if at least one parking circumstance image, including information on the unoccupied parking spaces, is acquired through at least one camera, performing (i) a process of instructing a first Convolutional Neural Network (CNN) to apply at least one first CNN regression operation to the parking circumstance image, to thereby calculate each of one or more estimated probability distributions on each of the decision points, and (ii) a process of instructing a second CNN to apply at least one second CNN regression operation to the parking circumstance image, to thereby generate estimated relational linear segment information on relationships among at least part of the decision points;
• and (II) instructing a loss layer to perform (i) a process of generating a first loss by referring to the estimated probability distributions and its corresponding Ground-Truth (GT) probability distributions, and training at least part of parameters in the first CNN by performing backpropagation using the first loss, and (ii) a process of generating a second loss by referring to the estimated relational linear segment information and its corresponding GT relational linear segment information, and training at least part of parameters in the second CNN by performing backpropagation using the second loss.

A testing device for detecting one or more unoccupied parking spaces for testing by using one or more probability distributions for testing on one or more decision points for testing of the unoccupied parking spaces for testing and relational linear segment information on relationships among the decision points for testing, comprising:

• at least one memory that stores instructions;
• and
• at least one processor configured to execute the instructions to perform processes of:
• (I) on condition that (1) a learning device, if at least one parking circumstance image for training, including information on one or more unoccupied parking spaces for training, is acquired through at least one camera for training, has performed (i) a process of instructing a first Convolutional Neural Network (CNN) to apply at least one first CNN regression operation to the parking circumstance image for training, to thereby calculate each of one or more estimated probability distributions for training on each of decision points for training, and (ii) a process of instructing a second CNN to apply at least one second CNN regression operation to the parking circumstance image for training, to thereby generate estimated relational linear segment information for training on relationships among at least part of the decision points for training;
• and (2) the learning device has instructed a loss layer to perform (i) a process of generating a first loss by referring to the estimated probability distributions for training and its corresponding Ground-Truth (GT) probability distributions, and training at least part of parameters in the first CNN by performing backpropagation using the first loss, and (ii) a process of generating a second loss by referring to the estimated relational linear segment information for training and its corresponding GT relational linear segment information, and training at least part of parameters in the second CNN by performing backpropagation using the second loss, performing (i) a process of instructing the first CNN to apply the first CNN regression operation to a parking circumstance image for testing, to thereby calculate each of one or more estimated probability distributions for testing on each of the decision points for testing, and (ii) a process of instructing the second CNN to apply the second CNN regression operation to the parking circumstance image for testing, to thereby generate estimated relational linear segment information for testing on relationships among at least part of the decision points for testing;
• (II) performing (i) a process of instructing a first selection layer to generate estimated decision point location information for testing on at least part of the decision points for testing, by referring to the estimated probability distributions for testing, and (ii) a process of instructing a second selection layer to generate estimated linear segment location information for testing on at least part of pairs for testing of at least part of the decision points for testing by referring to the estimated relational linear segment information for testing;
• and (III) instructing a pairing layer to detect the unoccupied parking space for testing by referring to the estimated decision point location information for testing and the estimated linear segment location information for testing.