Vehicular Vision Object Tracking via Temporal Attribute Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle vision systems using fisheye lenses face challenges in accurate object representation and classification due to distortion and variations in object size, form, and orientation, leading to errors in detection and classification, especially with temporal inconsistencies.
Innovation Solution
A vehicular vision system that includes a post-processing step using an assignment operator and temporal filtering to improve detection and classification accuracy by smoothing outputs and merging attributes of objects, employing techniques like the Hungarian algorithm and Kalman filter to handle variations and missed detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If fisheye lenses are used in vehicle vision systems to capture wide-field images, then the field of view is improved, but object representation accuracy deteriorates due to distortion and variations in object size, form, and orientation
Solution Approach 1:
The system dynamically adjusts detection parameters and object attributes based on the detected object's position in the image frame. Objects in different regions (center vs. periphery) have their size, form, and orientation parameters adjusted according to the fisheye distortion model, compensating for the lens distortion and maintaining accurate object representation across the entire wide field of view.
2Speed
If object detection is performed on each frame independently, then detection speed is improved, but temporal consistency deteriorates leading to errors in detection and classification
Solution Approach 1:
The system performs preliminary detection on the current frame, then uses assignment operators to match detected objects with tracked objects from previous frames before final classification. This preliminary matching step ensures temporal consistency by maintaining object identity across frames while preserving detection speed through efficient assignment algorithms.
Solution Approach 2:
The system implements feedback loops where detection results from previous frames are fed into the current frame processing. Tracked object positions, sizes, and attributes from prior frames serve as feedback to guide current frame detection and classification, improving temporal consistency while maintaining detection speed through iterative refinement.
3Reliability
If tracking is performed to maintain object identity across frames, then temporal consistency is improved, but computational complexity increases
Solution Approach 1:
The system extracts only the essential attributes needed for tracking (position, size, orientation) and uses assignment operators to match these extracted features across frames. By focusing on key discriminative features rather than full image processing, the system maintains temporal consistency while reducing computational complexity.
Solution Approach 2:
The system creates simplified copies of object representations (bounding boxes, key attributes) from previous frames and uses these copies for matching and tracking purposes. This copying approach maintains object identity across frames with minimal computational overhead, avoiding the need to process full high-resolution images for tracking.
Data Source
AI summary
A vehicular vision system includes a camera disposed at a vehicle and capturing image data. The vehicular vision system, via processing at an electronic control unit of a first frame of image data captured by the camera, detects a first object exterior of the vehicle and determines an attribute of the first object. The vehicular vision system, via processing at the electronic control unit of a second frame of image data captured by the camera, detects a second object exterior of the vehicle and determines an attribute of the second object. The system determines whether the second object is part of the first object. The vehicular vision system, responsive to determining that the second object is part of the first object, merges the attribute of the first object with the attribute of the second object and tracks the first object in subsequent frames of captured image data.