Video Recognition Feature Aggregation by Object and Frame
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recognition technologies aggregate feature vectors from all joint points in a moving image without distinguishing frames or objects, leading to potential errors in recognition due to unrelated joint points being associated, thereby decreasing recognition accuracy.
Innovation Solution
A recognition device that extracts individual feature quantities from videos of varying unit sizes and aggregates them for each unit, using neural networks with permutation-equivariant characteristics to ensure accurate recognition by generating aggregated feature quantities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If all feature vectors from all joint points are aggregated without distinction of frames and objects, then the aggregation process is simple, but unrelated joint points are associated leading to decreased recognition accuracy
Solution Approach 1:
The patent segments the video into multiple unit images (frames or objects) and performs aggregation separately for each unit. The extraction unit extracts feature vectors from joint points, and the aggregation unit aggregates these features for each individual unit image rather than all images together. This segmentation prevents unrelated joint points from being incorrectly associated while maintaining manageable processing complexity through systematic organization.
2Productivity
If feature vectors are aggregated for the entire moving image, then processing efficiency is high, but the accuracy of recognition decreases due to erroneous associations
Solution Approach 1:
The patent divides the video processing into separate aggregation operations for each unit image. The aggregation unit processes feature vectors frame-by-frame or object-by-object, which reduces erroneous associations. The recognition unit then uses these individually aggregated features to improve recognition accuracy while maintaining reasonable processing efficiency through the structured approach.
Solution Approach 2:
The patent applies different aggregation treatments to different local units (frames or objects) in the video. Each unit image receives individual aggregation processing tailored to its specific content, allowing the system to maintain high recognition accuracy by preserving local characteristics while still achieving overall processing efficiency through consistent methodology.
3Device complexity
If aggregation is performed without distinguishing unit images, then the system structure is simple, but unrelated joint points are associated causing erroneous recognition results
Solution Approach 1:
The patent introduces a clear segmentation structure where the extraction unit identifies unit images and the aggregation unit processes each unit separately. This structured segmentation maintains reasonable system complexity while dramatically improving reliability by preventing the association of unrelated joint points across different frames or objects.
Data Source
AI summary
To provide a recognition device that suppresses a decrease in recognition accuracy. A recognition device that performs recognition processing on a video obtained by capturing includes a neural network 172 that extracts, from a video including a plurality of pixels having a size of a first unit and a plurality of objects having a size of a second unit larger than the size of the first unit and smaller than a size of an entire video, an individual feature quantity indicating a feature of the pixel having the size of the first unit, a MaxPooling unit 173 that aggregates, in a case where a plurality of individual feature quantities are extracted, the plurality of extracted individual feature quantities for each object having the size of the second unit, and a DNN unit 178 that recognizes an event appearing in the video on the basis of an aggregation result.