Multi-Object Tracker with Occlusion and Hijacking Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking systems face challenges in accurately tracking multiple objects due to occlusion and hijacking, where objects of similar appearance can cause errors in identification and tracking.
Innovation Solution
A system and method that includes an image capturing unit and a multi-object tracker capable of detecting occlusion and hijacking by analyzing video frames, correcting the tracking image by removing occluded or hijacked objects, and determining depth order to prioritize the target object, using a reference background image for object extraction and tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple objects are tracked simultaneously using conventional object tracking systems, then tracking coverage is improved, but tracking accuracy deteriorates due to occlusion and hijacking
Solution Approach 1:
The system segments the tracking problem by separating detection and tracking functions. Detection identifies all objects in each frame, while tracking maintains identities across frames. This segmentation allows multiple objects to be tracked simultaneously without confusion, as each object has its own detection results and tracking history independently managed.
Solution Approach 2:
The system introduces depth information as an additional dimension to resolve occlusion and hijacking problems. By estimating depth order of objects, the system can determine which objects are in front of others, preventing incorrect tracking associations when objects of similar appearance overlap or pass near each other.
2Quantity of substance
If objects of similar appearance are tracked together, then tracking quantity is improved, but identification accuracy deteriorates due to hijacking
Solution Approach 1:
The system uses feedback mechanisms where detection results from current frames are continuously compared with tracking results from previous frames. When discrepancies are detected (such as sudden appearance of similar objects or unexpected displacements), the system adjusts tracking associations to maintain accuracy, allowing simultaneous tracking of multiple similar objects without hijacking.
Solution Approach 2:
Depth estimation adds a new dimension for distinguishing between similar objects. By determining the depth order, the system can differentiate objects that appear similar in 2D image space, enabling accurate identification and tracking of multiple objects with similar appearances simultaneously.
3Duration of action of stationary object
If occluded objects are tracked, then tracking continuity is improved, but tracking reliability deteriorates
Solution Approach 1:
The system performs preliminary detection of all objects in each frame before performing tracking association. This preliminary detection ensures that even occluded objects are identified, and their potential trajectories are established in advance, allowing the tracking system to maintain continuity and reliability by correctly associating detections with existing tracks even during occlusion events.
Solution Approach 2:
Depth information serves as a preliminary indicator for occlusion detection. By estimating depth order, the system can identify which objects are occluding others, allowing it to maintain reliable tracking by focusing on visible portions and using depth cues to predict occluded object positions, thus maintaining both continuity and reliability.
Data Source
AI summary
The present invention relates to a system for tracking an object. The system includes an image capturing unit configured to capture a video of a predetermined observation area and output the captured video; and a multi-object tracker configured to output an object-tracking image by tracking multiple objects within an object image which is generated by extracting the objects from each of image frames obtained from the video obtained from the image capturing unit, wherein the multi-object tracker determines whether occlusion of the objects or hijacking occurs while performing multi-object tracking, and when it is determined that at least one of the occlusion and hijacking occurs, the multi-object tracker outputs the object-tracking image corrected by removing the occurring occlusion or hijacking.