Image Segmentation via Temporal Motion Trajectories

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods struggle to robustly detect and segment moving persons with deformation, especially in crowded scenes, due to difficulties in extracting candidate regions and the need for a large number of parameters for articulated objects, leading to increased calculation complexity and errors.

Innovation Solution

An image processing method that analyzes motions of blocks in images, calculates temporal motion trajectories, and projects distances into a nonlinear space using infinitized, threshold-based, or exponentially transformed geodesic distances, followed by dimensionality reduction and clustering to segment articulated objects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If candidate regions of physical objects are extracted from images and parameterized models are applied, then object detection and segmentation become possible, but in crowded scenes with objects coming and going, proper extraction of candidate regions becomes difficult and detection robustness deteriorates

Engineering Contradiction:
Improveobject detection robustnessVSAvoidcandidate region extraction difficulty
Core Design Contradiction:
ReliabilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent extracts motion information and temporal trajectories from video sequences, separating the dynamic motion patterns from static background and other objects. By focusing on temporal motion characteristics rather than spatial candidate region extraction, the method avoids the difficulty of extracting candidate regions in crowded scenes while maintaining detection robustness

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transitions from analyzing spatial image data to analyzing temporal motion trajectories across multiple frames. By projecting motion data into a different dimensional space (temporal dimension) and applying nonlinear dimensionality reduction, the method enables robust object detection without relying on difficult candidate region extraction in crowded spatial scenes

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If parameterized physical object models are applied to extracted candidate regions, then object detection and segmentation become possible, but for articulated objects like persons with various postures, the vast number of parameters required induces increased calculation amount and model application errors

Engineering Contradiction:
Improveobject detection accuracyVSAvoidmodel parameter complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts essential motion characteristics (temporal trajectories and motion patterns) from video sequences, isolating the key dynamic features that define object identity and behavior. By focusing on motion patterns rather than full parameterized models, the method reduces computational complexity while maintaining detection accuracy for articulated objects

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the problem from using a vast number of parameters to describe articulated object postures to using a smaller set of motion trajectory parameters. By changing the parameter representation from detailed geometric models to temporal motion patterns, the method reduces calculation amount while improving robustness for objects with varying postures

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2418622B1Image processing method and image processing apparatus
Publication Date: 2015.12.02 PANASONIC HOLDINGS CORP
  • EP2418622B1 patent drawingFigure 1
  • EP2418622B1 patent drawingFigure 2
  • EP2418622B1 patent drawingFigure 3

AI summary

To provide an image processing apparatus which robustly performs segmentation on an image including an object such as a moving person with its deformation. The image processing apparatus includes: an image inputting unit (101) which receives temporally successive images; a motion analyzing unit (102) which calculates motions of blocks using at least two temporally different images and calculates, based on the motions of the blocks, temporal motion trajectories of the blocks in the temporally successive images; a distance calculating unit (103) which calculates a distance which indicates a similarity of the motions of the blocks, using a temporal motion trajectory of a block i and a temporal motion trajectory of a block other than the block i calculated by the motion analyzing unit; and a nonlinear space processing unit (104) which projects the distance calculated by the distance calculating unit into a nonlinear space and performs the segmentation on a result of the projection in the nonlinear space.