Method and device for extracting typical target material attributes based on spectral video data

Abstract

The invention provides a typical target material attribute extraction method and device based on spectral video data. The method comprises the following steps: collecting RGB-IR spectral video images;transforming RGB-using pre-trained regression model Wherein the IR image is a characteristic spectrum segment image; extracting reflectivity characteristics from each pixel value of the characteristic spectrum segment image; using a pre-trained classification model to classify the reflectivity characteristics of the characteristic spectrum segment image, judging whether each pixel value belongs to one of a plurality of materials or not, and giving a material attribute label graph; and processing the material attribute label map by using morphological opening operation and filling operation, and outputting the material attribute label map. According to the method, the reflection characteristic difference of different materials is utilized, the illumination-independent reflectivity characteristics are extracted, the reflectivity characteristics are judged by using the classifier, and the attribute extraction of the specific material in video monitoring is realized.

Description

technical field [0001] The invention relates to a spectral video processing method and device, in particular to a material attribute extraction method and device based on spectral video data, belonging to the field of video monitoring. Background technique [0002] At present, the video surveillance system has spread all over the city and has been widely used. The existing video surveillance system collects RGB video images, extracts the apparent features in the RGB images, uses the apparent features to extract and analyze targets, and realizes functions such as target detection and target tracking. But the appearance characteristics are affected by the video image quality. The images collected by the same target in different shooting environments, different postures, occlusions, etc. are very different, resulting in a great change in the apparent characteristics of the target, which is difficult to process and analyze. As a result, the existing system has low operating ef...

AI Technical Summary

Problems solved by technology

[0004] However, traditional material attribute extraction methods and systems are difficult to meet the needs of video surveillance, making it difficult to apply to video surveillance

First, the imaging equipment used in the traditional material attribute extraction method cannot shoot video and cannot be applied to video surveillance

The traditional method uses scanning imaging equipment to collect high-precision spectra of the target, and the imaging time is long, which makes it impossible to collect spectral videos

Second, the traditional method has high requirements on the shooting scene

In video surveillance scenarios, illumination changes are more common, and traditional reflectance features cannot be extracted.

Third, video surveillance is faced with the impact of large imaging noise

The exposure time of video imaging equipment is short, and the noise in the image is large, while the scanning equipment used in the traditional method has small noise, so the traditional system and method do not consider special processing of imaging noise

The above situation makes the traditional material attribute extraction method unable to be directly applied to video surveillance

Images