Apparatus for detecting objects in multi-directional images based on deep neural network and method therefor

摘要

In the method for detecting an object in a multi-directional image, when a main image and a sub image photographed in different directions are input while the feature extraction module shares one axis with respect to the same object, the input main image and sub image respectively generating at least one main image feature map and at least one sub image feature map by extracting features; generating, by an attention module, an attention map using the main image feature map and the sub image feature map; generating a feature map by combining the main image feature map and the attention map by a combining module; and detecting, by an object detection module, an object in the main image from the feature map.