Behavior Recognition Model Using Cumulative Feature Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based behavior recognition models face challenges in processing video clips with varying behavior times, leading to increased processing time and reduced accuracy due to repeated calculations and mismatched frame speeds.
Innovation Solution
The method involves generating sampling frames at a first sampling interval, creating cumulative feature maps by accumulating skeleton feature points, and learning a behavior recognition model using these maps, which allows for behavior inference across video clips with different behavior times by storing feature points in a queue to eliminate repeated calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-based models process video clips with varying behavior times using traditional methods, then behavior recognition can be performed, but processing time increases and accuracy decreases due to repeated calculations
Solution Approach 1:
The system pre-extracts and stores skeleton feature points from video clips in a queue before behavior recognition is needed. This preliminary processing allows the recognition model to access pre-computed features directly, eliminating repeated calculations during actual behavior recognition and reducing processing time while maintaining accuracy.
Solution Approach 2:
The system creates a queue that stores copies of extracted skeleton feature points from multiple video clips. These copied feature points can be directly utilized during behavior recognition without re-extracting them, thereby reducing computational redundancy and processing time while preserving recognition accuracy.
2Ease of manufacture
If the system processes video clips with different behavior times using fixed sampling intervals, then frame extraction is simplified, but feature point mismatch occurs leading to reduced recognition accuracy
Solution Approach 1:
The system dynamically adjusts the sampling interval for each video clip based on its specific behavior time. Instead of using a fixed sampling interval, the system calculates an optimal sampling interval for each clip, allowing flexible frame extraction that adapts to varying behavior durations. This dynamic approach maintains recognition accuracy by ensuring appropriate frame density for each specific behavior while keeping the extraction process simple.
Data Source
AI summary
An electronic device for performing behavior recognition and an operation method thereof are provided. The method of operating the electronic device includes generating sampling frames for each video clip by performing sampling on a plurality of video clips at a first sampling interval, generating a cumulative feature map for each video clip based on the generated sampling frames for each video clip, and using the cumulative feature map for each video clip as an input, learning a behavior recognition model for determining a behavior of an object included in a target video clip, wherein the plurality of video clips may include the object performing a same behavior, and wherein behavior time, which represents time consumed from a start to an end of the same behavior performed by the object, may be different for each video clip.