Video target detection method and device based on sparse foreground prior, and storage medium

Abstract

The invention discloses a video target detection method based on sparse foreground prior, a storage medium and equipment. A foreground extraction method based on orthogonal subspace learning is adopted to calculate and obtain a sparse foreground prior graph corresponding to each frame in a video; utilizing the ResNet feature extraction network and the feature pyramid structure to obtain semantic enhancement feature maps of the video frame and the sparse foreground image thereof; cascading the semantic enhancement feature map of the sparse foreground prior map with the semantic enhancement feature map of the current frame, and performing convolution fusion operation to obtain foreground prior fusion features of the current frame; mapping on each pixel of the foreground prior fusion featuremap to generate a candidate anchor box; and inputting the foreground prior fusion features and all anchor boxes into the trained classification and regression sub-network to obtain the category and position coordinates of the target object. The sparse foreground prior of the video data is fully mined, and the target detection accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a video target detection method, storage medium and equipment based on sparse foreground prior. Background technique [0002] Computer vision is an important field of artificial intelligence that trains computers to learn and understand real-world vision. With the help of pictures and videos and deep learning models, it is possible to accurately classify and identify the targets of interest, and then make further judgments and processing. Computer vision is generally divided into main tasks such as image recognition, object detection, and instance segmentation. Among them, the classification task generally gives the content description of the entire picture, while the detection task pays more attention to specific objects of interest, requiring both recognition and positioning results of the object of interest. Compared with the classification task, detectio...

AI Technical Summary

Problems solved by technology

However, there are new challenges in applying image detection techniques directly to video detection tasks

First, applying the image object detection network directly to each frame in the video for detection will bring huge computational costs; second, traditional image object detection methods cannot effectively utilize the priors of video data temporal continuity and sparse foreground, Difficult to mine timing characteristics in video data

Images