A learning method and a learning device for improving a neural network supporting autonomous travel by performing a sensor fusion integrating information acquired through a radar capable of distance prediction and information acquired through a camera Test method and test device using the same

摘要

To provide a method for learning a CNN using a camera and a radar together so that the CNN works suitably, even when the object depiction rate of a photographed image that is acquired through a camera is low due to that a photographing condition is unsuitable.SOLUTION: The method includes: the learning device causing convolution computation to be applied to a multichannel integrated image by a convolution layer and generating a feature map (S01); causing output computation to be applied to the feature map by an output layer and generating predictive object information (S02); causing a loss to be generated by a loss layer using the predictive object information and original correct answer object information corresponding thereto and at least some of the internal parameters of the CNN to be learned by carrying out back propagation using the loss (S03).SELECTED DRAWING: Figure 3