A Method of Feature Extraction and Target Tracking Based on Convolutional Neural Network

Abstract

The invention discloses a feature extraction and target tracking method based on a convolutional neural network. First, the performance of the network in feature extraction and foreground segmentation is improved through an offline pre-training method; and then the first frame of the calibrated video sequence is put into Into the network for online training, fine-tuning the parameters of the network model, so as to improve the processing ability of the convolutional neural network in specific problems. By adding a random two-dimensional mask and multiple iterations, it not only improves the accuracy of network prediction, but also avoids the problem of overfitting. The optionality of multiple scales also enables this method to adaptively select the scale of the target during the target tracking process. In terms of network parameter update, the update is performed in real time by setting a threshold, which improves the accuracy and robustness of the target tracking method.

Description

technical field [0001] The invention relates to the field of computers, and more particularly, to a feature extraction and target tracking method based on a convolutional neural network. Background technique [0002] Object tracking problem is an important part in the field of computer vision. Its basic task is to give a video sequence, and to calibrate the initial state of the target (such as position, size, etc.) in the first frame, and automatically estimate the state of the target in subsequent frames through a series of algorithms. Current tracking algorithms can be divided into two types: generative model and discriminative model. The generative method uses a generative model to describe the apparent characteristics of the target and searches for candidate targets to minimize the reconstruction error. Representative algorithms such as sparse coding, online density estimation, principal component analysis ( PCA, principal component analysis), etc.; the discriminative ...

AI Technical Summary

Problems solved by technology

The main reason is that the training set provided in target tracking is too small, usually only contains the initial state calibrated in the first frame of the video sequence, it is difficult to give full play to the advantages of deep learning methods, the efficiency and accuracy of deep learning methods in tracking, etc. There is still a lot of room for improvement

Images