Abnormal moving target tracking system and method based on deep neural network

Abstract

The invention relates to the field of transaction target tracking, and provides a transaction target tracking system and method based on a deep neural network, and the system comprises an image collection module, a frame image feature extraction module, and a deep neural network model construction module. The method comprises the following steps: acquiring positive sample images of different types of transaction targets in a specified scene, and acquiring negative sample images corresponding to the positive sample images according to the types of the transaction targets and the positions of the transaction targets in the specified scene; according to the collected positive and negative sample images, extracting features of a previous frame of negative sample image for discovering a transaction target, features of multiple frames of positive and negative sample images when the transaction target appears, and features of a first frame of negative sample image after the transaction target disappears; and constructing and training a deep neural network model according to the extracted features of each frame of image, and automatically identifying whether a moving target appearing in the specified scene is a transaction target and the type of the transaction target through the trained deep neural network model.

Description

technical field [0001] The invention relates to the field of moving target tracking, in particular to a moving target tracking system based on a deep neural network. Background technique [0002] Certain scenarios, such as border protection zones and military no-passing zones, often require precise monitoring of any unattended target. As an important confidential or no-go zone, there are often illegal intrusion investigations, and of course, there are also no-no-go zones. Therefore, it is extremely important to accurately identify and track moving targets in specific scenarios such as restricted areas. [0003] With the rapid development of Internet technology, the monitoring methods for areas are becoming more and more mature. For example, the patent application with the application number: CN201910353400.3, which discloses a kind of human action based on a convolutional neural network tracking a specific target in a monitoring scene The identification method firstly acqui...

AI Technical Summary

Problems solved by technology

However, this patent application solves the problems of low frequency of abnormal actions in the monitoring scene, difficulties in data collection and labeling, and the large number of characters and complex actions in the monitoring scene. What is the type of target, and there is no relevant solution for it

Images