Video Object Re-Identification Thresholds for Scene Exit Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional re-identification methods in object tracking systems risk incorrectly identifying similar but different objects as the same due to visually similar appearances, especially in scenarios where objects exit and re-enter the scene, leading to loss of tracking information.
Innovation Solution
Adjust the re-identification threshold to require greater similarity between objects for re-identification, allowing continued tracking of objects that have exited and potentially re-entered the scene, while maintaining tolerance for temporary occlusions and camera view changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the re-identification threshold is set to a lower value to allow re-identification of similar objects, then the tracking continuity is improved, but the risk of misidentifying different objects as the same object increases
Solution Approach 1:
The re-identification threshold is made dynamic rather than static. The system adjusts the threshold based on whether an object has exited the scene: using a first (lower) threshold for objects within the scene to maintain tracking continuity, and a second (higher) threshold for re-identifying objects that have re-entered to prevent misidentification. This dynamic adjustment resolves the contradiction by adapting the threshold to the specific tracking context.
Solution Approach 2:
The system changes the re-identification threshold parameter based on the tracking state. When an object exits and re-enters the scene, the threshold is adjusted from a first value to a second value, which is configured to prevent misidentification. This parameter change allows the system to maintain tracking continuity for occluded objects while preventing false re-identification of similar but different objects.
2Measurement precision
If the re-identification threshold is increased to prevent misidentification, then the object identification accuracy is improved, but the tracking continuity for occluded objects deteriorates
Solution Approach 1:
The system dynamically switches between two threshold values based on the tracking scenario. A first threshold maintains tracking continuity during occlusions, while a second threshold prevents misidentification after scene exit. This dynamic approach ensures that high accuracy is achieved when needed without sacrificing tracking continuity during normal occlusions.
Solution Approach 2:
The re-identification threshold parameter is changed based on the object's tracking history. When an object has exited the scene, the threshold is set to a higher value to prevent misidentification. This parameter change ensures that identification accuracy is maintained for re-entering objects while allowing more flexible matching for continuously tracked objects.
3Productivity
If tracking is terminated when an object exits the scene to save computational resources, then the computational efficiency is improved, but the ability to track re-entering objects deteriorates
Solution Approach 1:
The system performs preliminary action by adjusting the re-identification threshold in advance before an object re-enters the scene. When an object exits, the threshold is adjusted to a second value that prevents misidentification. This preliminary adjustment ensures that when the object re-enters, the system can correctly identify it as a new object rather than incorrectly matching it with the previous object, thus maintaining reliable re-entry detection.
Solution Approach 2:
The system uses feedback from the object exit event to adjust the re-identification threshold. When an object exits the scene, this feedback triggers a threshold adjustment that affects subsequent re-identification attempts. This feedback mechanism ensures that computational resources are used efficiently while maintaining the ability to correctly detect and track re-entering objects.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to an aspect, there is provided a computer-implemented method of tracking objects in a video sequence of a scene, the method comprising: determining a location of a sink in the scene where objects exit the scene and a location of a source where objects enter the scene; tracking a first object moving in the scene using a re-identification algorithm, wherein the first object is associated with a re-identification threshold of the re-identification algorithm; detecting that the first object has exited the scene at the sink; and responsive to detecting that the first object has exited the scene at the sink, adjusting the re-identification threshold associated with the first object such that a probability that the re-identification algorithm re-identifies a second object, entering the scene at the source after the first object has exited the scene at the sink, as the first object is reduced.