Object Tracking Using Segmented Feature Vectors for Occlusion Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking technologies face difficulties in accurately identifying and tracking objects due to the occlusion phenomenon, which results in low accuracy and incorrect identification of object identities, especially when only half of the object's information is available.
Innovation Solution
An electronic apparatus and method that generate and compare whole and partial object vectors using a specific object model and a database of reference vectors to improve object re-identification accuracy and recall rates by considering the correlation between multiple frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing object detector is used to generate object detection bounding box, then object detection can be performed, but object feature vector loses half of the original information and is error-prone under occlusion
Solution Approach 1:
The patent divides the object feature extraction into multiple segmentation ratios (e.g., 0.5, 0.6, 0.7, 0.8, 0.9). Instead of relying on a single detection bounding box that may lose information under occlusion, the system generates multiple feature vectors at different segmentation levels. This allows the system to capture both partial and complete object features, thereby reducing information loss and improving identification accuracy even when the object is partially occluded.
2Reliability
If existing object tracking technology is used, then object tracking can be performed, but the accuracy of marking the correct identity of the object is not high when tracking the object under occlusion
Solution Approach 1:
The patent performs preliminary action by generating and storing multiple object feature vectors at different segmentation ratios in advance (before tracking). These pre-computed feature vectors are stored in a database with their corresponding segmentation ratios. During tracking, when occlusion occurs, the system can directly retrieve and compare these pre-computed vectors, avoiding the need to re-extract features from occluded images and thereby maintaining high tracking accuracy.
Solution Approach 2:
The patent changes the parameter of segmentation ratio to multiple discrete values (0.5, 0.6, 0.7, 0.8, 0.9) instead of using a single fixed ratio. This parameter change allows the system to adapt to different occlusion levels by selecting appropriate segmentation ratios for comparison, thereby improving reliability under varying occlusion conditions.
3Device complexity
If only half of the object information is available from detection bounding box, then tracking process can be simplified, but identification becomes error-prone and difficult
Solution Approach 1:
The patent adds another dimension to the tracking process by introducing segmentation ratio as an additional comparison dimension. Instead of comparing only single feature vectors, the system compares multiple feature vectors across different segmentation ratios. This dimensional expansion allows the system to maintain simplified tracking operations while significantly improving identification accuracy by utilizing more comprehensive object information.
Data Source
AI summary
A method used for object tracking includes: using a specific object model to generate a first vector of a first ratio object and a second vector of a second ratio object of an image in an object detection bounding box of a specific frame; generating an identity label of an object within the bounding box according to the first vector, the second vector, and M first ratio reference vectors and M second ratio reference vectors stored in an object vector database.