Tiled Optical Flow Tracking for Long-Term Vehicle Image Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for linking optical flow across multiple images struggle with maintaining accuracy and availability over extended periods due to geometric changes in the image setting, leading to inaccuracies and reduced selectivity in object detection.
Innovation Solution
A method and device for concatenating flow vectors across a series of images by subdividing a track map into tiles, extending flow tracks based on associatable flow vectors, and using a multi-index concept to ensure uniform distribution and reduce quantization errors, allowing real-time updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of moving object
If optical flow is linked across multiple images using conventional methods, then tracking duration is extended, but measurement precision deteriorates due to geometric changes in image setting
Solution Approach 1:
The track map is subdivided into multiple tiles, and flow vectors are processed and extended within individual tiles rather than across the entire map. This segmentation allows maintaining precision within each tile while enabling long-term tracking by progressively extending tracks across multiple tiles over time.
2Duration of action of moving object
If flow tracks are extended across multiple images, then tracking duration increases, but device complexity increases due to multi-index management
Solution Approach 1:
A multi-index system is introduced that adds temporal and spatial dimensions to the flow track data structure. Instead of simple linear tracking, the system uses multiple indices to reference flow vectors across different images and tiles, enabling efficient long-term tracking without proportionally increasing processing complexity.
3Measurement precision
If flow vectors are densely distributed, then object detection selectivity improves, but quantization errors increase
Solution Approach 1:
Different regions of the track map are treated with different levels of flow vector density and quantization precision. The system adapts the distribution and processing quality of flow vectors based on local requirements, allowing high selectivity in critical regions while managing quantization errors through localized processing strategies.
Data Source
AI summary
A method for linking optical flow across a plurality of images of an image detection device for a vehicle. At least one tile is selected from a track map, the track map including a plurality of flow tracks and being subdivided into a plurality of tiles. At least one flow track encompassed by the tile is subsequently extended, using a flow vector of a flow map that is associatable with the flow track. The flow map includes a plurality of flow vectors that are based on a present image of the plurality of images.