Remote multi-target efficient detection and identification method

Abstract

The invention discloses a remote multi-target efficient detection and identification method. The method comprises the following steps that S1, a terminal processes a video stream into a picture set through a current limiting algorithm; S2, original pictures of the picture set are normalized into sample images with fixed sizes; S3, with pixel points of the sample images as nodes, the state of an undirected graph is depicted by using an adjacent matrix of all the nodes; S4, an s latitude feature vector of each node is obtained; S5, a complete vectorized tag dictionary is pre-constructed, and predicted values of tags are obtained from the output feature matrix through the dictionary; S6, the terminal carries out threshold screening on the predicted values of the tags; S7, a server generates a corresponding hash code according to a tag result, and enters different target recognition model channels through the hash code; and S8, the server recognizes all known targets in advance and then stores the recognized targets into a feature vector library, effective pictures enter corresponding model channels and then are analyzed into feature vectors, a result is obtained after the feature vectors are compared with the feature vector library, and the result is fed back to the terminal.

Description

technical field [0001] The invention relates to the technical field of computer vision and pattern recognition, in particular to a method for efficient detection and recognition of remote multi-targets. Background technique [0002] At present, target recognition is one of the hot research issues in the field of computer vision and pattern recognition. How to reduce server computing load and avoid network congestion has become a key and difficult issue in target recognition. [0003] In the early days of the development of artificial intelligence, computers can easily deal with some problems that are difficult or even impossible for humans to solve. These problems can be described by a formal mathematical law. The real tasks faced by artificial intelligence are those tasks that are difficult to describe with formal symbols. Since the deep neural network algorithm first shined on the ImageNet dataset, the field of object detection has gradually begun to use deep learning for ...

AI Technical Summary

Problems solved by technology

[0005] Nowadays, most of the remote target detection is used in the remote monitoring system. The traditional remote recognition method is in the target detection and recognition at the edge end. The front end collects information and transmits it to the server through the network. The server uses algorithm matching and then returns the recognition result to the termi

Images