Information tracking method based on convolutional neural network

Abstract

The invention provides an information tracking method based on a convolutional neural network. The method comprises the following steps: constructing a correlation filtering convolutional neural network in combination with correlation filtering and a convolutional neural network; constructing a time flow convolutional neural network and a spatial flow convolutional neural network on the basis; the three parts are constructed to form a deep network in a jumping connection mode; training the deep network until the three models are all converged; respectively extracting image block feature information in the current frame and a motion change feature information set of a target between frames in a plurality of time sequences through a time stream convolutional neural network and a spatial stream convolutional neural network; fusing the image block feature information and the motion change feature information weight, constructing a full-connection neural network, and obtaining prediction information of the current frame target; and fusing all models by using a Bagging algorithm to determine final prediction information of the current frame, constructing a time network and a space network on the basis of a related filtering network, further capturing time information and space information of a target, and improving the accuracy of the algorithm.

Description

technical field [0001] The present invention relates to the technical field of tracking methods, in particular to an information tracking method based on a convolutional neural network. Background technique [0002] Visual target tracking task has been a research hotspot in the field of computer vision and has been widely used, especially in recent years with the rapid development of scientific and technological productivity, video surveillance, UAV flight, automatic driving and other fields are in urgent need of excellent target tracking algorithms; visual target The tracking task describes that in a given video sequence scene, only the position of the target in the first frame is provided, and then the algorithm predicts the next position and size of the target. Although a large number of algorithms have emerged in recent years, there is still no Better to solve this task, because there are great challenges, especially target appearance deformation, scale change, object oc...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above technical problems, the present invention provides an information tracking method based on a convolutional neural network to solve the problem that the existing information tracking method does not fully utilize the temporality of the video, does not grasp the motion information between the tracking target frames and only uses the last The problem that one layer of features is not enough to represent the target

Images