Self-adaptive depth perception visual relation detection method

According to the adaptive depth perception visual relation detection method, a depth perception visual fusion module is composed of a depth refining unit and a depth perception attention unit, and inaccurate depth visual information and accurate RGB visual information can be fused. Through regional local pooling operation, when a large error exists in depth space positioning of a target under the influence of background noise and foreground noise, a self-adaptive depth space positioning method is adopted, and information correlation of each small region in a bounding box is measured by using regional information variance; depth information in two main factors of visual relation detection, namely visual appearance and spatial position, is fully utilized, and the performance of the netting meridians and collaterals in a visual relation detection task is improved. 本发明的自适应深度感知视觉关系的检测方法，由深度感知视觉融合模块由深度细化单元和深度感知注意力单元组成，能够将不准确的深度视觉信息与准确的RGB视觉信息融合。通过分区域局部池化操作，在背景噪声和前景噪声的影响下，目标的深度空间定位存在较大的误差时，采用自适应深度空间定位方法，利用区域信息方差度量边界