Method for training object classification model and identification method using object classification model

Abstract

The invention discloses a method for training an object classification model. The method comprises the following steps of: acquiring a plurality of frames of images, and determining initial foreground object areas from the images; performing principal axis normalization conversion on the initial foreground object areas in the images of different frames; aligning the corresponding initial foreground object areas subjected to the principal axis normalization conversion in the images of different frames to acquire a final foreground object area; extracting characteristics from the final foreground object area; and training the object classification model by using the extracted characteristics. By the principal axis normalization conversion of the extracted foreground object areas, the method overcomes the influence of different visual angles on the shapes of the foreground object areas and reduces intra-class distance; and meanwhile, the foreground object areas on the images of different frames are aligned by adopting block matching technology, and the extracted effective foreground object area in the foreground object areas is used as the final foreground object area.

Description

【Technical field】 [0001] The invention relates to the technical field of pattern recognition, in particular to the training of an object classification model and a recognition method using the model. 【Background technique】 [0002] Object type recognition is an important research topic in the field of intelligent video surveillance, and it is the application of computer image processing technology and pattern recognition technology in the field of intelligent video surveillance. Object type recognition technology is widely used in road monitoring, financial banking, waterway management and other industries. [0003] In the existing object type recognition technology, there are many different methods of object type recognition technology, such as: the object type recognition technology based on heuristic rules, which uses simple information such as the shape, size, and proportion of the object to extract Foreground objects are analyzed to obtain the object type. This method ...

AI Technical Summary

Problems solved by technology

[0012] Recently, some people use block Local Binary Pattern (LBP) to calculate sample features in this object type training and recognition method. Although the block LBP represents the neighborhood information of the object to a certain extent, this method is used in Some images can also achieve good results, but this method has not made good use of inter-frame information

On the other hand, in the existing object type training and recognition methods, the extraction of the foreground object area is simply zoomed in, and the influence of different viewing angles in the image on the object features is not considered.

Images