Rapid acquisition and calibration method of image data set in deep learning

摘要: The invention discloses a rapid acquisition and calibration method of an image data set in deep learning. After electronic stabilization processing is carried out on collected video, denoisingand smoothening are by median filtering, foreground extraction performed using Gaussian mixture model; according to the position outer profile extracted foreground, corresponding information original video frame located obtain minimum bounding box specific target automatically; classification iscarried hog-feature-based local template matching algorithm; with inter-frame algorithm, same-class marking connection region between adjacent frames; type obtained realize set. Therefore, problems that only few carry tags learning new difficulty high cost solved; great convenience provided for training identification data.