Lens shielding discrimination method based on semantic analysis

Abstract

The invention provides a lens occlusion discrimination method based on semantic analysis, and belongs to the technical field of image recognition. According to the technical scheme, the lens occlusion judgment method based on semantic analysis comprises the steps that a deep convolution detection network is trained, monitoring areas and occlusion objects are detected, the monitoring areas and the multiple occlusion objects are paired into multiple semantic relation pairs in a one-to-many mode, the sum of relation pair detection frames formed by combining subjects and objects is a relation external connection frame, and the sum of the relation pair detection frames is the relation external connection frame. And mapping to a top feature map of the backbone convolutional neural network, deducting the feature map by using a ROIAlign algorithm, mapping to a fixed size, and sending the feature map to a semantic discrimination full-connection layer for relation prediction. And finally, predicting and judging whether the current lens is shielded or not according to whether the prediction result contains shielding or not. The beneficial effects of the invention are that the method is used for decomposing the classification problem of lens shielding into a detection / relation prediction problem, is higher in interpretability, and can accurately judge the position relation between a shielding object and a monitoring area.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a method for judging lens occlusion based on semantic analysis. Background technique [0002] At present, the control camera is the core device used in the construction of safe cities and safe communities. As time goes by, the environment where the cameras were installed in the early years may have changed a lot, and some are even blocked by (such as dense branches and leaves). Thousands of cameras are inspected manually one by one, which is laborious and may be missed due to fatigue of personnel. Ordinary classification or detection algorithms, camera lens occlusion detection related algorithms are very few, indicating that it is not easy for a computer to judge this problem. Currently, only a few algorithms detect whether the lens is occluded by detecting the presence of leaves. This method has two limitations: first, there is no necessary connection between the foc...

AI Technical Summary

Problems solved by technology

However, it is difficult to judge the lens occlusion by target detection and image classification alone.

Images