Object Tracking Map Generation for Video Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object tracking in video processing becomes computationally expensive when dealing with multiple tracking targets or high target density, as it requires extensive comparisons of feature amounts near predicted moving destinations.
Innovation Solution
An image processing apparatus that detects objects, generates a map representing object correspondences across multiple images, and matches objects based on this map, utilizing convolutional neural networks (CNNs) for robust feature integration and matching, reducing processing costs by leveraging multiple map types such as line segment, intermediate point, distance, and ID maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional feature amount comparison methods are used for object tracking, then tracking accuracy can be maintained, but processing cost increases significantly when dealing with multiple tracking targets or high target density
Solution Approach 1:
The patent segments the tracking process into distinct functional modules: a detection unit that identifies objects in captured images, a generation unit that creates correspondence maps between detected objects, and a determination unit that performs matching based on these maps. This segmentation allows each module to specialize in a specific task, improving overall processing efficiency while maintaining tracking accuracy through coordinated operation of the components.
Solution Approach 2:
The generation unit creates correspondence maps in advance that represent relationships between objects across multiple images. By pre-establishing these spatial and temporal relationships before the actual matching operation, the determination unit can perform efficient comparisons without needing to conduct exhaustive feature amount comparisons, thereby reducing processing cost while preserving tracking reliability.
2Measurement precision
If exhaustive feature amount comparisons are performed near predicted moving destinations, then object matching accuracy improves, but computational complexity increases
Solution Approach 1:
The correspondence map serves as an intermediary data structure that encodes spatial and temporal relationships between objects. Instead of directly comparing feature amounts between all possible object pairs, the system uses this intermediate representation to guide the matching process, significantly reducing computational complexity while maintaining matching accuracy through the structured information contained in the maps.
Solution Approach 2:
The patent transforms the tracking problem from direct feature amount comparison into a map-based correspondence problem. By changing the representation parameters from raw feature vectors to structured correspondence maps containing spatial relationships, the system enables more efficient matching operations that reduce computational complexity while preserving the information needed for accurate matching.
Data Source
AI summary
There is provided with an image processing apparatus. A detection unit detects an object from a captured image. A generation unit generates a map representing a correspondence between objects detected in a plurality of captured images. A determination unit matches the objects detected in the plurality of captured images based on the generated map.