Video Object Tracking via Background Model Adaptation for Occlusion Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems are ineffective in tracking static objects over time and handling occlusion situations, where objects are partially or fully obscured by other objects, leading to difficulties in maintaining accurate object tracking.
Innovation Solution
An object tracking apparatus and method that processes video frames to detect and track both moving and static objects by generating difference images, clustering objects, and updating background models, enabling continuous tracking even when objects are occluded and allowing for the identification of unattended objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current object tracking methods are used to separate objects from background and determine motion vectors, then moving objects can be tracked, but static objects cannot be tracked for lengthy periods and occlusion situations cannot be handled
Solution Approach 1:
The system dynamically adapts its tracking approach based on object characteristics. For static objects, it uses background subtraction with learned background models; for moving objects, it uses motion vector analysis. This dynamic adaptation allows the system to maintain tracking continuity across different object types and situations, resolving the contradiction between reliable tracking and versatility.
Solution Approach 2:
The tracking system is designed to handle multiple object types (static and moving) and multiple scenarios (occlusion and non-occlusion) using a unified framework. The system can switch between different tracking strategies depending on the situation, making it universally applicable to various surveillance scenarios while maintaining reliable tracking continuity.
2Productivity
If traditional motion-based tracking is used, then moving objects can be detected, but static objects cease to be tracked after a short interval
Solution Approach 1:
The system performs preliminary background learning and creates background models before actual tracking begins. This preliminary action establishes a reference framework that enables continuous tracking of static objects without requiring motion detection, thereby extending the tracking duration for stationary objects while maintaining efficient detection capabilities.
Solution Approach 2:
The system dynamically switches between motion-based detection for moving objects and background-subtraction-based detection for static objects. This dynamic approach allows the system to maintain high productivity for moving object detection while simultaneously achieving extended tracking duration for static objects.
3Measurement precision
If objects are tracked by separating them from background via motion vectors, then moving objects can be followed, but occluded objects cannot be tracked
Solution Approach 1:
The system uses background models as an intermediary reference framework that persists even when objects are occluded. When an object is occluded, the background model provides continuous spatial context, allowing the system to maintain tracking reliability by predicting object position based on historical data and background structure, rather than relying solely on direct object-background separation.
Solution Approach 2:
The system performs preliminary learning of background structures and object characteristics before occlusion occurs. This preliminary action creates a knowledge base that enables the system to maintain accurate object location measurements even during occlusion events, by using the pre-learned information to fill in gaps when direct observation is unavailable.
Data Source
AI summary
An apparatus and method for the analysis of a sequence of captured images covering a scene for detecting and tracking of moving and static objects (86) and for matching (88) the patterns of object behavior in the captured images to object behavior in predetermined scenarios.