Shadow-Assisted Object Tracking for Occluded Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video object tracking is complex due to fast-moving objects, occlusions, and changes in orientation, which complicates the detection and association of objects across video frames, especially in scenarios like sporting events where occlusions and multiple players obscure each other.
Innovation Solution
The method utilizes shadow information from objects to improve object recognition and tracking by segmenting, localizing, and extracting 3D information, including the use of dominant shadow directions to accurately determine object positions and poses, even in occluded conditions, by analyzing shadow projections and their intersections to refine object location estimates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional background subtraction and feature matching techniques are used for object tracking, then the system can operate with standard video processing, but tracking accuracy deteriorates in occluded scenarios and when objects move fast relative to frame rate
Solution Approach 1:
The patent segments the object into multiple parts by tracking extreme points (corners, tips, ends) rather than treating the object as a whole. This segmentation allows each extreme point to be tracked independently, improving accuracy during occlusions and fast motion while maintaining manageable system complexity through focused feature tracking.
Solution Approach 2:
The patent introduces shadow information as an intermediary element to assist object tracking. By detecting and analyzing shadows cast by objects, the system gains additional spatial and temporal information that helps maintain tracking accuracy during occlusions and fast motion, without requiring complex multi-camera systems.
2Reliability
If object recognition techniques are added to video tracking to handle fast motion and orientation changes, then tracking robustness improves, but processing time increases due to the complexity of analyzing multiple features across frames
Solution Approach 1:
The patent applies partial action by tracking only extreme points rather than all features of an object. This selective approach provides sufficient tracking robustness for fast motion and orientation changes while significantly reducing processing time compared to full object recognition techniques.
Solution Approach 2:
The system performs preliminary detection of extreme points in each frame before attempting to associate them with previous frames. This preliminary identification simplifies the subsequent association process and reduces overall processing time while maintaining tracking robustness.
3Productivity
If multiple objects are tracked simultaneously, then the system provides comprehensive scene monitoring, but association difficulty increases when objects are close together or occluding each other
Solution Approach 1:
The patent segments the tracking problem into independent extreme point tracking tasks. By treating each extreme point as a separate tracking entity, the system can monitor multiple objects simultaneously while simplifying the association process, as extreme points provide distinctive features that are easier to match even when objects are close or occluded.
Solution Approach 2:
Shadow detection serves as an intermediary that provides additional spatial information to resolve ambiguities in object association. When multiple objects are present and occluding each other, shadow information helps distinguish between them and correctly associate extreme points with their source objects.
4Adaptability or versatility
If the tracked object changes orientation and pose over time, then the tracking system can handle more dynamic scenarios, but the complexity of maintaining accurate object models increases
Solution Approach 1:
The patent applies partial action by focusing tracking on extreme points rather than maintaining complete object models. This approach handles dynamic scenarios with changing orientation and pose effectively while avoiding the complexity of full object modeling, as extreme points provide sufficient information for tracking without requiring detailed pose estimation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Described herein are apparatus, systems and methods for shadow assisted object recognition and tracking. The methods performed by the apparatus and system include identifying a blob within a video image, the video image having at least one object and at least one shadow of the at least one object, the at least one shadow of the at least one object cast by at least one light source. Identifying the blob includes identifying an object projection corresponding to the at least one object and a shadow projection corresponding to the at least one shadow. A location of an object portion of the at least one object is determined based on the shadow projection.