Object Attention Maps for Precise Video Event Feature Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video analysis techniques struggle to easily detect features attributable to specific objects in videos, such as games, due to the complexity of multiple objects and events.

Innovation Solution

An image analysis system utilizing a machine learning model to generate attention maps and tokens for specific objects, combined with an event prediction mechanism, to facilitate easier detection of features and events in video content.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If visual observation or related methods are used to detect important events in videos, then events can be detected, but it is not easy to detect features attributable to a specific object due to multiple objects in the video

Engineering Contradiction:

Improvefeature detection accuracyVSAvoidobject-specific feature detection difficulty

Core Design Contradiction:

Measurement precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The patent segments the video analysis task by generating separate attention maps for each object type (player, ball, goal). This segmentation allows the system to focus on object-specific features independently, resolving the difficulty of detecting features attributable to specific objects among multiple objects in the video.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces attention maps as an intermediary between the input video and event detection. These attention maps highlight regions associated with specific objects, serving as a mediator that enables precise detection of object-specific features without being overwhelmed by multiple objects in the scene.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Adaptability or versatility

If multiple objects are present in video images, then rich video content is captured, but detecting features focused on a specific object becomes more difficult

Engineering Contradiction:

Improvevideo content coverageVSAvoidobject-specific feature detection accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system segments the analysis by creating distinct attention maps for different object types (player, ball, goal). This allows the system to maintain versatility in handling multiple objects while achieving precision in detecting features for each specific object type through dedicated attention mechanisms.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by generating attention maps with different characteristics for different object types. Each attention map is optimized to highlight features specific to its target object type, enabling the system to adapt to multiple objects while maintaining high detection accuracy for each specific object.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS12555378B2Image analysis system, image analysis method, and program

Publication Date: 2026.02.17 SONY INTERACTIVE ENTERTAINMENT LLC

AI summary

Disclosed herein is an image analysis system including a machine learning model configured to receive input of an image and to output an image feature quantity and a map source for image analysis, a map generation section configured to generate, on the basis of a vector corresponding to an object and the output map source, an attention map indicative of a region associated with the object in the image, and a token generation section configured to generate a token indicative of a feature associated with an event of the object on the basis of the generated attention map and the image feature quantity.