Object Tracking via Body Part Segmentation in Overlapping Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques struggle to accurately detect and track individuals in crowded scenes where objects overlap significantly, leading to difficulties in identifying multiple persons and maintaining accurate tracking.
Innovation Solution
An image processing apparatus and method that includes a processor configured to detect object regions, estimate the number of objects within these regions, and track objects using the estimated counts, employing a detecting unit, estimating unit, and tracking unit to improve accuracy in overlapping environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If silhouette-based detection methods are used to detect persons, then detection speed is improved, but detection accuracy deteriorates in crowded scenes where persons overlap
Solution Approach 1:
The patent segments the detection process into multiple specialized detection units (head detection unit, shoulder detection unit, leg detection unit) that independently detect different body parts. This segmentation allows the system to maintain high detection speed while improving accuracy in crowded scenes by detecting multiple discrete body parts rather than relying on a single silhouette contour that may be obscured by overlaps.
2Device complexity
If traditional object detection methods are used, then implementation simplicity is maintained, but tracking reliability deteriorates when objects largely overlap
Solution Approach 1:
The patent implements dynamic tracking by continuously updating object information across multiple frames and using movement patterns to maintain tracking. The tracking unit dynamically adjusts to object movements and uses temporal information from sequential frames to maintain reliable tracking even when objects overlap, transforming a static detection problem into a dynamic solution that adapts to changing scenes.
Solution Approach 2:
The patent incorporates feedback mechanisms where detection results from multiple body parts are integrated to improve overall object detection and tracking reliability. The system uses feedback from head, shoulder, and leg detections to confirm and refine object tracking, creating a self-correcting system that maintains reliability in overlapping conditions.
3Productivity
If single-object detection approaches are used, then processing efficiency is maintained, but the ability to detect multiple overlapping persons deteriorates
Solution Approach 1:
The patent segments the person detection task into detection of multiple independent body parts (head, shoulders, legs) that can be processed in parallel. This allows the system to maintain processing efficiency by using independent detection units while simultaneously increasing the number of detectable objects, as each body part can be detected and counted separately even when persons overlap.
Solution Approach 2:
The patent transitions from detecting single objects to detecting multiple body parts across different spatial dimensions and locations within the image. By detecting head, shoulder, and leg regions at different positions, the system effectively adds dimensional complexity to the detection process, enabling it to identify multiple overlapping persons while maintaining processing efficiency through parallel operation of detection units.
Data Source
AI summary
A technique of tracking objects with high accuracy even in an environment where the objects largely overlap with each other is provided. An image processing apparatus includes a detecting unit configured to detect an object region from an input image, an estimating unit configured to estimate a number of objects included in the detected object region, and a tracking unit configured to track objects included in the object region using the estimated number of objects.