Object detection method in occlusion conditions

Abstract

Object detection method of the present invention includes a data augmentation model (10) for detecting the state of a fruit or fruits wherein the model (10) comprises the steps of: detecting a currently processed bunch bounding box; verifying if a system-generated random number in the range of 0 to 1 exceeds a predetermined hyperparameter. If not, the process goes back to the step of detecting a currently processed bunch bounding box; if it exceeds, the process continues to generate an occlusion region. The process further comprises detecting if the occlusion region contains one or more target bunch bounding boxes. When it contains zero target bunch bounding box, the process comprises moving the occlusion region directly onto the currently processed bunch bounding box. When it contains one or more target bunch bounding boxes, the process comprises generating a new coordinate for a generated region bounding box and moving the occlusion region onto the currently processed bunch bounding box; in the meantime, recalculating the new coordinate of the generated region bounding box. After all the currently processed bunch bounding boxes in the input image have been processed, generating an augmented output image. Figure. 1a