Global feature and local feature-based cross-lens target retrieval system and method

Abstract

The invention discloses a global feature and local feature-based cross-lens target retrieval system and method. The method comprises the steps of obtaining a foreground target image from a monitoring video; preprocessing the foreground target image, and extracting a color feature, a texture feature and the like with a plurality of feature dimensions from the image; and after dimension reduction processing, storing the features in a database, so that the target retrieval system is formed. When the target retrieval system is used for retrieving a query image, firstly feature information of the query image is extracted, and the query image and the foreground target image in the target system are subjected to global feature similarity calculation to obtain a primary screening result; and secondly local feature information of the query image is extracted, the query image and the foreground target image subjected to primary screening are subjected to local feature similarity calculation, and the foreground target image with high global and local feature similarity is taken as a query result image. According to the system and the method, the identification accuracy and performance of cross-lens target retrieval can be improved.

Description

technical field [0001] The invention relates to a cross-shot target retrieval system and method based on global features and local features, belonging to the technical field of image processing and pattern recognition. Background technique [0002] With the development of video surveillance technology, in order to achieve the efficiency of retrieving objects from massive surveillance videos, cross-camera object retrieval technology emerges as the times require. Query the target, greatly shorten the time to find the original video, and improve the efficiency and recognition accuracy. There are two main methods for cross-border head target retrieval: [0003] One is to extract a variety of features with stability (consistent features at different times) and distinctiveness (inconsistent features at the same time or different times) from the original image, and splicing multiple features in series to form the feature vector of the target image. . The problem with this method...

AI Technical Summary

Problems solved by technology

The problem with this method is that it is necessary to collect a large number of original images of the same target from different cameras as samples, the collection workload is large, there are many types and models of cameras, and it is difficult to realize, and the two target images that need to be matched are shot by different cameras, resulting in There are various differences between pictures, resulting in differences in the matrices obtained by different camera training. Using only one metric matrix is ​​not enough to project features in two different modalities into the same space for distance calculation.

Images