Online learning method for video target detection

Abstract

The invention discloses an online learning method for video target detection, and belongs to the field of machine vision. The method comprises the steps of model pre-training, target detection, tracking correction, label correction, key frame extraction and model iterative updating. The method comprises the steps of firstly, training a current model of a basic version by utilizing an open source or self-annotation data set; pre-detecting the video sequence by using the current model; utilizing an improved KCF tracking algorithm and a k-nearest neighbor algorithm based on a feature space to respectively perform box correction and label correction on a pre-detection result; extracting a video key frame by utilizing a key frame extraction method based on feature space similarity measurement,and removing repeated images; and training the model by utilizing the key frame data and the correction detection result to realize iterative updating of the model. According to the method, detectionand labeling results are corrected through a tracking algorithm and clustering analysis, a target detection model is retrained by utilizing the corrected results, continuous improvement of model performance is realized, and self-adaptation of an application scene is realized.

Description

technical field [0001] The invention belongs to the fields of deep learning and machine vision, and in particular relates to an online learning method for video target detection. Background technique [0002] Target detection is to find out the object of interest in the image, including two sub-tasks of object positioning and object classification. It is one of the basic tasks in the field of machine vision and has a wide range of applications in the fields of intelligent transportation, intelligent manufacturing, security monitoring, and automatic driving. . With the development of deep learning, target detection algorithms have gradually shifted from traditional algorithms based on manual features to deep learning algorithms based on neural networks. The current research on target detection mainly focuses on two directions: image-based target detection and video-based target detection. [0003] The image-based target detection algorithm was first proposed, which is divid...

AI Technical Summary

Problems solved by technology

Therefore, whether it is based on images or videos, deep learning algorithms have domain adaptation problems, that is to say, it is difficult to achieve full-domain detection through a single target detection model

When the scene changes, in order to make the detector achieve a certain accuracy, it is often necessary to re-collect data, manually label, and then re-train the model, which requires a lot of manpower and time.

Images