Detection Propagation for Robust Real-Time Multiple-Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MOT techniques face challenges such as high computational intensity, reduced detection performance, and failures due to changing object appearance, occlusions, and sensor artifacts, particularly in tracking diverse object classes.
Innovation Solution
The MOT-P framework implements a two-stage detection matching process with a detection propagation unit and end-to-end per-class hyperparameter optimization, allowing for extrapolation of detections without continuous object detection, thereby maintaining high processing speeds and improving tracker performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If end-to-end tracking is used to provide better consistency to tracks, then track consistency is improved, but computational resources increase and detection performance decreases
Solution Approach 1:
The tracking system is segmented into two distinct components: end-to-end tracking for high-priority objects and tracking-by-detection for other objects. This segmentation allows each component to operate optimally for its intended purpose without the computational overhead of applying both methods uniformly to all objects.
Solution Approach 2:
Different tracking qualities are applied to different objects based on their priority classification. High-priority objects receive the computationally intensive end-to-end tracking with better consistency, while lower-priority objects use the faster tracking-by-detection method, optimizing resource allocation locally for each object type.
2Productivity
If tracking by detection is used for fast processing, then processing speed is improved, but track consistency deteriorates
Solution Approach 1:
The system segments the tracking workload by object priority, applying tracking-by-detection only to lower-priority objects where speed is more critical than perfect consistency, while reserving end-to-end tracking for high-priority objects that require better track consistency.
Solution Approach 2:
Instead of applying full end-to-end tracking to all objects, the system applies partial tracking (tracking-by-detection) to objects where full tracking would be excessive, achieving acceptable results with reduced computational effort for non-critical objects.
3Measurement precision
If object detection is performed on every frame, then detection accuracy is improved, but frame rate falls below threshold
Solution Approach 1:
The system performs preliminary classification of objects by priority before determining the detection strategy. High-priority objects are pre-identified for end-to-end tracking which maintains accuracy, while lower-priority objects are pre-marked for detection propagation to maintain frame rate.
Solution Approach 2:
Instead of continuous detection on every frame for all objects, the system uses periodic detection only for high-priority objects when needed, while relying on detection propagation for lower-priority objects, creating an intermittent detection pattern that maintains accuracy where needed while preserving frame rate.
4Productivity
If detection propagation is used to maintain frame rate, then processing speed is improved, but detection accuracy decreases
Solution Approach 1:
Different detection qualities are applied locally to different objects: high-priority objects receive accurate end-to-end detection when frame rate permits, while lower-priority objects use detection propagation with acceptable accuracy reductions, optimizing the overall system performance.
Solution Approach 2:
The system applies partial detection (detection propagation) to objects where full detection accuracy is not critical, accepting reduced accuracy as a reasonable trade-off for maintaining frame rate, while reserving full detection accuracy for high-priority objects.
Data Source
AI summary
A data processing system implements obtaining a frame of video content at an object detection pipeline, the video content comprising a plurality of frames; analyzing the frame using an object detection model to detect a plurality of objects and associate each object with a confidence score; performing a primary matching operation on high confidence detection objects to associate the high confidence detection objects with an object track of a plurality of object tracks, the high confidence detection objects being objects associated with a confidence score that satisfies a confidence threshold; performing a secondary matching operation on low confidence detection objects to associate the low confidence detection objects with an object track of the plurality of object tracks, low confidence detection objects being objects associated with a confidence score that does not satisfy the confidence threshold; and outputting the plurality of object tracks.