REINFORCEMENT LEARNING-BASED LABEL-FREE SIX-DIMENSIONAL ITEM ATTITUDE PREDICTION METHOD AND DEVICE

摘要

A reinforcement learning-based label-free six-dimensional item attitude prediction method and device. The method comprises: obtaining a target image (14) for prediction, the target image (14) being a two-dimensional image (201) comprising a target object; performing, according to the target image (14), attitude prediction using a pre-trained attitude prediction model (12), so as to obtain a prediction result (16), the attitude prediction model (12) being a model (202) obtained by performing reinforcement learning according to a sample image; and determining, according to the prediction result (16), a three-dimensional position and a three dimensional direction (203) of the target object. The attitude prediction model (12) is trained by introducing reinforcement learning, and attitude prediction is performed according to the target image (14) by using the pretrained attitude prediction model (12), so that the problem of six-dimensional item attitude estimation based on a two-dimensional image can be solved when there is no real attitude labelling, and the prediction effect of label-free six-dimensional item attitude prediction is ensured.