Object Tracking via Limb Convex Hull Rotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In computer stereo vision, particularly for intelligent driving, accurately tracking objects like pedestrians or animals in city scenes is challenging due to posture changes, background variations, and distance changes, making it difficult to maintain object tracking effectively.
Innovation Solution
A system and method for tracking objects in a disparity map using a processor and storage that executes computer-executable instructions, involving modules like convex hull extraction, torso detection, limb matching, and template updating to align principal gradient directions, enabling accurate object detection and tracking by comparing and rotating limb convex hulls to match prior and current frame features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional object tracking methods are used in disparity maps, then the tracking process can be performed, but the tracking accuracy deteriorates due to posture changes, background variations, and distance changes
Solution Approach 1:
The object is segmented into multiple body parts (torso, head, limbs) and each part is tracked independently using convex hull extraction. This segmentation allows the system to handle posture changes more effectively by tracking individual body parts rather than treating the entire object as a single entity.
Solution Approach 2:
The patent transforms the tracking problem from 2D image space to 3D disparity space by extracting convex hulls in the disparity map. This dimensional transformation provides depth information that helps distinguish objects from background variations and improves tracking robustness.
2Measurement precision
If detailed object features are extracted for accurate tracking, then tracking precision improves, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential geometric features (convex hulls) of body parts from the full image data. By taking out only the necessary shape information and discarding redundant pixel data, the system achieves accurate matching with reduced computational complexity.
Solution Approach 2:
The patent replaces complex image processing operations with geometric convex hull extraction and matching. This substitution simplifies the computational process while maintaining tracking accuracy by focusing on shape geometry rather than full image analysis.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Disclosed is an object tracking method including steps of extracting prior convex hulls in an object template of a prior frame, which includes at least a prior torso convex hull and prior limb convex hulls; detecting a portion corresponding to the prior torso convex hull, in a current frame as a current torso portion; extracting current limb convex hulls from a surrounding region of the current torsos portion in the current frame; for each of the prior limb convex hulls, rotating, based on principal gradient directions of the current and prior limb convex hulls, each of the current limb convex hulls so as to let its principal gradient direction be consistent with the principal gradient direction of the corresponding prior limb convex hull; seeking rotated current limb convex hulls matching the prior limb convex hulls, respectively; and determining, based on the matching results, an object template in the current frame for achieving object tracking.