OBJECT RECOGNITION METHOD, OBJECT RECOGNITION DEVICE, AND CLASSIFIER TRAINING METHOD

摘要

Disclosed is an object recognition method comprising steps of obtaining an original image containing a region of an object waiting for recognition as well as depth information and an original optical flow image corresponding to the original image; utilizing the depth information to conduct a compensation process with respect to the original optical flow image so as to obtain a compensated optical flow image; using a convolutional neural network to extract feature components at least in the compensated optical flow image; and taking advantage of a classifier to carry out classification with respect to the feature components extracted so as to acquire a classification result for recognizing the object.