Behavior Recognition Based on 3D Point Cloud and Key Skeletal Nodes

A 3D and node technology, which is applied in the field of behavior recognition based on 3D point cloud and key skeletal nodes, can solve the problems of ignoring geometry and topology, low recognition rate of complex behavior, difficult trade-off between real-time processing and high-precision recognition, etc. Achieve the effect of improving the accuracy of behavior recognition and improving the accuracy of recognition

Active Publication Date: 2019-06-25
HUNAN UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

These methods have the following three defects: first, they are sensitive to lighting, camera angles, background changes and partial occlusion; second, they ignore the inherent geometry and topology of the behavior itself; Difficult to identify trade-offs
However, most of the methods for extracting features from depth maps are similar to RGB, sensitive to changes in camera angles and actor speeds, and have low recognition rates for complex behaviors involving human-object interactions.

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
  • Behavior Recognition Based on 3D Point Cloud and Key Skeletal Nodes
  • Behavior Recognition Based on 3D Point Cloud and Key Skeletal Nodes
  • Behavior Recognition Based on 3D Point Cloud and Key Skeletal Nodes

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0025]Action Recognition on the MSR-Action 3D Dataset. The MSR-Action 3D data set contains 20 behaviors, namely: high arm swing, cross arm swing, hammering, grasping, forward boxing, high throwing, drawing X, drawing hook, drawing circle, clapping hands, swinging with both hands , Side boxing, bending, forward kicking, side kicking, jogging, tennis swing, tennis serve, golfing, picking up, and throwing, each act is performed 2 or 3 times by 10 people. The actors collected in this database are in a fixed position, and most of the behaviors mainly involve the movement of the upper body of the actor. First, we directly extract the 3D point cloud sequence from the depth sequence, and divide the 3D point cloud sequence into non-overlapping (24×32×18) and (24×32×12) along the X, Y, and T directions respectively. spatio-temporal units; our method is then tested using cross-validation, i.e. five subjects for training and the remaining five for testing, exhaustively 252 times. Table ...

Embodiment 2

[0027] Action Recognition on the MSR Daily Activity 3D Dataset. The data set contains 16 behaviors, consisting of 10 behavior subjects, each behavior subject performs the behavior 2 times, one standing, one sitting, a total of 320 behavior videos. The 16 behaviors are: drinking, eating, reading, talking on the phone, writing, sitting, using a notebook, vacuum cleaning, laughing, throwing paper, playing games, lying on the sofa, walking, playing guitar, standing, and sitting. The experimental setup is the same as above, this database is extremely challenging, not only contains intra-class variation, but also involves human-object interaction behavior. Table 2 is a comparison of the recognition rates of different methods on this database. It can be seen from the table that our method has achieved an accuracy rate of 98.1%, and the average accuracy rate has reached 94.0±5.68%, which is an excellent experimental result.

Embodiment 3

[0029] Action Recognition on the MSR Action Pairs 3D Dataset. This data set is a data set of behavior pairs, including 12 behaviors and 6 groups of behavior pairs, namely: pick up a box, put down a box, lift a box, place a box, push a chair, pull a chair , put on a hat, take off a hat, carry a backpack, take off a backpack, stick a poster, pull a poster. In this database, there are similar motion and shape cues between each behavior pair, but their temporal correlation is opposite. The experimental settings are the same as above. Table 3 is the comparison of all existing popular methods on this database. Our method has achieved a recognition rate of 97.2%.

[0030] Table 1: Performance of existing methods on the MSR Action 3D dataset. Mean±STD is calculated from 252 cycles. The 5 / 5 column means that {1,3,5,7,9} of the agents are used for training and the rest are used for testing.

[0031]

[0032] Table 2: Comparison of MSR Daily Activity recognition rates. Mean±STD i...

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 relates to a behavior recognition system based on 3D point cloud and key skeleton nodes. Specifically: extract the 3D point cloud sequence from the depth map, then evenly divide the 3D point cloud sequence into N non-overlapping spatiotemporal units, calculate the local position model (LPP) of each spatiotemporal unit, and calculate the local position model statistics Deviation Descriptor (SDLPP). In addition, a subset of key skeletal nodes is extracted from 3D skeletal nodes using the node motion algorithm, and the 3D node position features of key skeletal nodes and the local occupancy model (LOP) in the corresponding depth map are calculated. Finally, the above three heterogeneous features are cascaded, and the stochastic determination forest is used to mine distinguishable features, classify, and identify behaviors. The invention extracts 3D local geometric features and dynamic time features of human behavior, has a high recognition rate for complex human behavior involving interaction between human and objects, and is suitable for complex human behavior recognition.

Description

technical field [0001] The invention belongs to the fields of artificial intelligence and pattern recognition, and in particular relates to behavior recognition based on 3D point clouds and key skeletal nodes. Background technique [0002] Human behavior recognition mainly refers to the analysis and recognition of the action types and behavior patterns of the observed people, and uses natural language to describe them. According to the complexity of behavior, some researchers divide human behavior into four levels: posture, individual behavior, interactive behavior, and group behavior. At present, most of the researches are still mainly focused on the first two levels, while there are relatively few research reports on the latter two levels. Human behavior recognition technology has broad application prospects and very considerable economic value. The application fields involved mainly include: video surveillance, medical diagnosis and monitoring, motion analysis, intellige...

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 Patents(China)
IPC IPC(8): G06K9/00
CPCG06V40/20
Inventor 张汗灵
Owner HUNAN 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