Image-based object recognition method and system based on learning of environment variable data

An image-based object recognition system, comprising:

• an image reception unit configured to receive an image acquired through one or more cameras and segment the image on a per-frame basis;
• a primary learning unit configured to primarily learn labeling results for one or more objects in the image segmented on a per-frame basis;
• a primary reasoning unit configured to perform object detection in the image through a model obtained from the primary learning and generate first boxes for regions in which objects are present in the image;
• a secondary learning unit configured to allocate a weight to labeled data according to reliability of the labeled data in the shape of each of the first boxes generated from the performing object detection and perform secondary learning for results of the performing object detection with weights allocated thereto;
• and
• a secondary reasoning unit configured to finally detect one or more objects in the image through a model generated from the performing secondary learning,
• wherein the secondary learning unit is further configured to calculate each of the weights to be applied to the respective first boxes based on values that are obtained based on an overlap area between a corresponding one of the first boxes obtained by the primary reasoning unit in the performing object detection and a second box most closely matching a truth, a distance between a centroid of the first box and a centroid of the second box, and an aspect ratio of the first box.