Visual semantics and position sensing method and system based on infrared thermal imaging

Abstract

The invention provides a visual semantics and position sensing method and system based on infrared thermal imaging. The method comprises the steps: acquiring an infrared thermal image through an infrared thermal imager; performing processing classification and instance segmentation on the image through a deep neural network; obtaining pixel coordinates of the ROI by using examples such as a minimum external matrix and an image moment; calibrating a matrix transformation relation from an image coordinate system to a camera coordinate system by using a visual calibration algorithm; performing normal distance measurement through distance measuring equipment; determining the category of the object in the image and the three-dimensional coordinates of the ROI; and transmitting data to an uppercomputer in a distributed node mode. According to the invention, the influence caused by environmental uncertainty is reduced, and stable and uniform light supplementing equipment does not need to beindependently installed. In addition, three-dimensional positioning is conducted on the object in the mode that the ultrasonic sensor and the depth camera are combined, and the positioning accuracy and stability are improved. And a reliable technical method is provided for rapid, accurate and robust visual identification and positioning.

Description

technical field [0001] The invention relates to the technical field of visual recognition, in particular to a visual semantics and position perception method and system based on infrared thermal imaging. Background technique [0002] Due to the development of artificial intelligence, visual recognition technology is more and more widely used in various fields. Visual recognition is to extract, process and understand the image information of objective objects, and finally use it for actual detection and control. [0003] The existing technical solution is to collect RGB images and classify them with OpenCV traditional examples or neural networks to identify related objects. For the category of visual semantic recognition, perform three-dimensional positioning. At present, 3D depth cameras are used to locate objects. The problem with this positioning method is that the point cloud images are not dense enough, and mirror objects are prone to reflection, making the positioning...

AI Technical Summary

Problems solved by technology

At present, 3D depth cameras are used to locate objects. The problem with this positioning method is that the point cloud image is not dense enough, and mirror objects are prone to reflection, making the positioning imprecise.

The difficulty of visual recognition in the prior art is that it is easily disturbed by the external environment, such as occlusion between objects, light and dark changes of objects caused by light, etc.

The existing technology uses RGB images that are easily affected by indoor and outdoor light; there is a big difference between daytime and nighttime imaging; special supplementary light equipment is required to stabilize the imaging

When the heat radiation object is blocked, such as when a person wearing a mask, a hat, etc. covers the face and head for imaging, it is impossible to accurately judge

Images