Object Tracking via Dual-Period Motion Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing tracking techniques in PTZ cameras suffer from a 'transfer problem' where a tracking object is erroneously determined as another object when the original object crosses another object, leading to incorrect image capture.
Innovation Solution
An image processing apparatus and method that utilize motion prediction based on short and long frame periods to accurately identify and track objects, even when they cross each other, by selecting appropriate identification information based on the prediction results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion prediction based on short and long frame periods is implemented, then tracking accuracy during object crossings is improved, but device complexity increases
Solution Approach 1:
The patent divides the motion prediction process into two distinct segments: short-frame-period motion prediction and long-frame-period motion prediction. Each segment uses different temporal resolutions to capture different aspects of object motion. The short-frame prediction handles rapid changes, while the long-frame prediction provides contextual stability, together resolving the tracking ambiguity during crossings without requiring a single complex prediction system.
Solution Approach 2:
The patent introduces a temporal dimension by utilizing multiple frame periods (short and long) to predict object positions. This multi-temporal approach adds a time-scale dimension to the tracking system, allowing it to distinguish between objects crossing at different speeds and patterns, thereby improving accuracy without proportionally increasing spatial or hardware complexity.
2Measurement precision
If dual-frame-period motion prediction is used, then object identification accuracy during crossings is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary motion predictions using both short and long frame periods before the actual object identification and crossing detection occurs. By pre-calculating expected object positions and trajectories using multiple temporal scales, the system prepares prediction data in advance, reducing the computational burden and processing time required during real-time crossing events.
Solution Approach 2:
The system performs motion prediction at two different frame periods (short and long), which represents a partial application of multi-temporal analysis focused on the critical crossing scenario. Rather than implementing exhaustive multi-scale analysis at all times, the patent applies the dual-frame-period prediction selectively and partially, balancing accuracy improvement with processing time constraints.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus identifies an object based on a predicted position of the object based on first frames in a first period and a position of the object in a target frame, identifies an object based on a predicted position of the object based on second frames in a second period and a position of the object in the target frame, and, in a case where objects are detected in the target frame, executes tracking of a tracking target object based on distances between the objects, and at least one of an identification result of the object based on a detected position of the object and the predicted position based on the first frames and an identification result of the object based on the detected position and the predicted position based on the second frames. The second period is longer than the first period.