Vehicle Object Tracking Using Multi-View Occlusion Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous driving systems face challenges in accurately tracking objects when they are obscured by other objects, leading to erroneous recognition and tracking failures.
Innovation Solution
The proposed solution involves an apparatus and method for controlling a vehicle that divides rear and front surface images and side surface images of objects, calculates their reliability, and analyzes the correlation between objects based on feature values of reliable images, using techniques such as Mahalanobis or cosine distance calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional camera-based object tracking is used, then the tracking method is simple, but the recognition accuracy deteriorates when objects are obscured by other objects
Solution Approach 1:
The patent segments the object tracking problem into multiple views by dividing images into front surface, rear surface, and side surface images. This segmentation allows the system to analyze objects from multiple perspectives, improving recognition accuracy when objects are obscured in any single view.
Solution Approach 2:
The patent transitions from two-dimensional single-view image analysis to multi-dimensional analysis by incorporating front, rear, and side surface images. This dimensional expansion enables the system to track objects even when they are obscured in any single view, significantly improving recognition accuracy.
2Measurement precision
If multi-view image analysis is implemented, then object recognition accuracy improves, but computational complexity increases
Solution Approach 1:
The patent applies local quality by calculating reliability values for different image regions (front surface, rear surface, side surface images) and selectively processing only the reliable portions. This approach improves recognition accuracy while reducing unnecessary computational overhead from processing low-quality or obscured regions.
Solution Approach 2:
The patent introduces reliability calculation as a new parameter to evaluate the quality of detected objects in each view. By changing the processing approach based on reliability parameters, the system optimizes computational resources by focusing on high-confidence detections and reducing processing of low-confidence regions.
3Reliability
If reliability calculation for different image surfaces is performed, then erroneous recognition is reduced, but processing time increases
Solution Approach 1:
The patent performs preliminary reliability calculations on front surface, rear surface, and side surface images before full object tracking processing. This preliminary action identifies and filters out low-reliability detections early, reducing the overall processing time by avoiding detailed analysis of questionable objects.
Solution Approach 2:
The patent introduces reliability scores as an intermediary metric between image detection and final object tracking decisions. This intermediary layer enables efficient filtering and prioritization of objects for detailed processing, improving tracking reliability while managing processing time through selective analysis.
Data Source
AI summary
An apparatus for controlling a vehicle includes a sensor having at least one camera to obtain information about objects positioned around the vehicle. The apparatus also includes a controller configured to detect at least one object image from an image obtained from the camera. The controller predicts a position of each object on the image currently obtained, based on information about objects recognized from an image previously obtained. The controller recognizes an object by analyzing correlation between two objects based on an object image detected at the predicted position and an object image previously recognized.