Compressed domain behavior recognition method based on multi-scale time sequence receptive field

A recognition method and receptive field technology, applied in the field of behavior recognition, can solve problems such as reducing algorithm speed

Active Publication Date: 2021-05-28
XI AN JIAOTONG UNIV +1
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The process of decompressing the original P frame greatly reduces the rate of the overall algorithm

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Compressed domain behavior recognition method based on multi-scale time sequence receptive field
  • Compressed domain behavior recognition method based on multi-scale time sequence receptive field
  • Compressed domain behavior recognition method based on multi-scale time sequence receptive field

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] The present invention is elaborated below in conjunction with accompanying drawing:

[0033] Such as figure 1 As shown, the present invention provides a compressed domain behavior recognition method based on multi-scale time-series receptive field, the following steps:

[0034] 1) Compressed domain data sampling, such as figure 2 As shown, for the input video stream in the compressed domain, the video is firstly divided into 8 segments at equal intervals, each segment randomly samples a video group, and each video group uses I-frame images and 4 P-frame motion vector images as input data .

[0035] 2) Multi-scale receptive field network structure, such as image 3 As shown, it contains two-way networks with long sequence and short sequence. For I-frame images and motion vector images, they are sent to the long-sequence receptive field network together. Through 2D-CNN and feature time-series movement, I-frame space and time-series features are extracted to obtain cl...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a compressed domain behavior recognition method based on a multi-scale time sequence receptive field, and the method comprises the steps: equally dividing a video into eight segments for a video stream after compression coding, randomly selecting a video group from each segment, decoding an I-frame motion vector image and four P-frame motion vector images of the video group, and carrying out the decoding of the I-frame motion vector image and the P-frame motion vector image; and sending to a multi-scale receptive field network comprising two branches, namely a long-time-sequence receptive field network and a short-time-sequence receptive field network, for prediction, wherein a final result is weighted average of prediction scores of the two branches. Meanwhile, a channel attention mechanism guided by motion vector features is added to a long-time-sequence receptive field branch, foreground features are highlighted, and background interference is reduced. Through training and testing on a public data set, the effectiveness of the method is verified, and the accuracy of compressed domain behavior recognition is obviously improved.

Description

technical field [0001] The invention belongs to the field of behavior recognition, and in particular relates to a compressed domain behavior recognition method based on multi-scale time series receptive fields. Background technique [0002] Behavior recognition technology realizes the classification of behaviors in videos by extracting discriminative motion features from videos. Difficulties in behavior recognition include large differences in the duration of different actions in the video, false detections caused by background interference when the background is complex, etc. The existing behavior recognition algorithms include image domain and compressed domain. The input of the image domain algorithm is the original unencoded video, while the compression domain algorithm is to analyze the video compressed and encoded by the video coding technology. [0003] With the development of urban intelligence, surveillance cameras in the city cover almost every corner of the city...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06K9/46G06K9/62G06N3/04G06N3/08H04N19/42H04N19/85
CPCG06N3/049G06N3/08H04N19/42H04N19/85G06V40/20G06V20/42G06V20/46G06V10/462G06F18/24
Inventor 李凡张斯瑾贺丽君
Owner XI AN JIAOTONG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap