Multi-Frame Video Action Recognition for Stable Gesture Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video gesture recognition systems face instability and smoothness issues due to uncontrollable factors such as complex backgrounds and non-standard actions.

Innovation Solution

A method for recognizing video actions by determining action categories and positioning information using a current video frame and at least one forward frame, incorporating a convolutional neural network and correction mechanisms to improve stability and smoothness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional video gesture recognition methods are used, then the system can process video frames, but the recognition results suffer from instability and smoothness issues due to complex backgrounds and non-standard actions

Engineering Contradiction:
Improvestability of recognition resultsVSAvoidcomplex background interference
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent segments the video processing into distinct phases: forward frame processing to predict current frame actions, current frame recognition, and backward frame correction. This segmentation allows each phase to focus on specific tasks, improving overall stability by handling complex background interference in controlled steps rather than attempting single-frame recognition

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary action by processing forward frames before the current frame to predict action categories and positioning information. This preliminary prediction provides a reference that guides the current frame recognition, enabling the system to anticipate and compensate for potential recognition errors before they occur, thereby improving stability

Inventive Principle:
Principle #10Preliminary action

2Reliability

If single-frame action recognition is used, then the processing speed is fast, but the recognition results lack smoothness and consistency

Engineering Contradiction:
Improvesmoothness of recognition resultsVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent implements periodic action by systematically processing frames in a repeating cycle: forward frames → current frame → backward frames → correction. This periodic multi-frame processing pattern ensures smoothness through temporal consistency while maintaining efficient processing节奏, balancing quality and speed

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

The system employs feedback mechanisms where backward frames are used to correct predictions made from forward frames. The correction process uses the actual current frame recognition results to adjust and refine the predicted action categories and positioning information, creating a closed-loop feedback system that improves smoothness without excessive time loss

Inventive Principle:
Principle #23Feedback

3Measurement precision

If action recognition is performed without correction mechanisms, then the processing is simple, but the recognition accuracy decreases under uncontrollable factors

Engineering Contradiction:
Improveaction recognition accuracyVSAvoidcomplexity of recognition system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces backward frames as an intermediary element that mediates between the forward frame predictions and the final recognition results. These backward frames serve as a corrective intermediary, adjusting the predictions to account for temporal context and reduce errors caused by complex backgrounds and non-standard actions, thereby improving accuracy without excessive complexity

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system replaces simple mechanical single-frame recognition with an intelligent multi-frame correction system that uses neural network-based prediction and correction. This substitution introduces sophisticated algorithms that can handle uncontrollable factors more effectively, improving measurement precision through intelligent processing rather than simple mechanical frame-by-frame analysis

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3862914B1Video action recognition method, apparatus, and device, and storage medium
Publication Date: 2026.03.11 BIGO TECH PTE LTD
  • EP3862914B1 patent drawingFigure 1~2
  • EP3862914B1 patent drawingFigure 3
  • EP3862914B1 patent drawingFigure 4a~4b

AI summary

Disclosed in the embodiments of the present application are a video action recognition method, apparatus, and device, and a storage medium. The method comprises: on the basis of a current video frame and at least one preceding video frame, determining an action type and action positioning information of the current video frame; and, on the basis of the action type and the action positioning information of the video frame, determining the action content of the video.