A dense connection convolutional network-based target tracking method

Abstract

The invention discloses a dense connection convolutional network-based target tracking method. The method comprises the following steps of S1, determining the size and position of an interested target; S2, extracting convolution features of an input frame and judging the convolution features, if the input frame is an initial frame, solving a PCA projection matrix to reduce the dimension of the convolution features, using the obtained convolution features to train a dense connection network-based target tracking model, entering S7, otherwise, using the trained PCA projection matrix to reduce the dimension of the convolution features of the input frame, and entering S3; S3, inputting the convolution features into a tracking model to predict the position of the target of interest; S4, performing scale sampling at the target prediction position, and estimating the size of the target; S5, updating the network weight of the target tracking model; S6, outputting a target prediction position and scale; and S7, inputting the next frame until all frames of the video are predicted. According to the invention, end-to-end learning of the tracking model is realized, the training time is effectively shortened, and the use efficiency is improved.

Description

technical field [0001] The invention relates to a target tracking method, in particular to a target tracking method based on a densely connected convolutional network, and belongs to the technical field of target tracking. Background technique [0002] Object tracking is an important research field in computer vision, which is widely used in security monitoring, unmanned driving, human-computer interaction and so on. The main purpose of object tracking is to estimate the motion state of a given object of interest in a video. As a hot topic, object tracking has achieved many outstanding research results in recent years. However, since the changes in illumination, changes in target appearance, and background occlusion during use will pose great challenges to target tracking algorithms, the research on target tracking algorithms still needs to be in-depth. [0003] In recent years, the target tracking algorithm based on correlation filtering has attracted the attention of man...

AI Technical Summary

Problems solved by technology

Since the pre-training model for extracting convolutional features and correlation filtering are independent of each other, it does not reflect the advantages of neural network end-to-end learning

At the same time, the boundary effect brought by cyclic sampling also seriously limits the performance of the correlation filter tracking algorithm.

In addition, in the training process of the above algorithm, a large amount of data needs to be used and a high cost of time is required, which is not convenient for daily application

Images