SURROUNDING SITUATION RECOGNITION METHOD FOR ACTIVELY DETECTING CHANGE IN OBJECT BY OVERLAPPING IMAGE-BASED OBJECT DETECTION AND SEMANTIC IMAGE SEGMENTATION

摘要

The present invention relates to a surrounding situation recognition method for actively sensing changes in objects by overlapping image-based object detection and semantic image segmentation. The method comprises the steps of: (a) applying a convolution technique to an input image to extract an encoded feature image; (b) applying a fully fonnected neural networks technique to the encoded feature image to detect a predetermined object region; and (c) changing the encoded feature image to divide the input image into at least one environment image region, wherein a change in correlation of the object region compared to the environment image region is sensed by counting the number of pixels mutually overlapping the object region or the environment image region, and using a change in the number of pixels. When the surrounding situation recognition method for actively sensing changes in objects by overlapping image-based object detection and semantic image segmentation proposed in the present invention is applied, there is a benefit of preventing danger or enabling swift response in the event of an accident since it is allowed to understand the correlation between an object in an image and a surrounding environment.;COPYRIGHT KIPO 2020