Video Object Fragmentation Detection via Spatial Representation Extension
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video object tracking methods face challenges with errors in object location extraction and tracking, particularly with partial or full occlusion, leading to incorrect merging of detections and failure in maintaining accurate object tracks.
Innovation Solution
A computer-implemented method that determines a current spatial representation for a detection in an image sequence by deriving an expected spatial representation from previous frames, extending it to obtain an extended spatial representation, and assessing similarity measures to classify detections as fragments of a video object, thereby improving the association of detections with tracks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple detections within a certain distance are treated as the same object, then detection failures are reduced, but over-merging of coincidentally close objects occurs
Solution Approach 1:
The system performs preliminary actions by deriving an expected spatial representation from previous frames before making merging decisions. This expectation, based on historical track data, is used to pre-filter and guide the association of current detections, preventing both missed merges and over-merging by establishing a predicted reference position in advance
Solution Approach 2:
The system uses feedback mechanisms by comparing the spatial representation of current detections with the expected spatial representation derived from previous frames. This feedback loop allows the system to adjust tracking decisions based on the discrepancy between expected and actual detection positions, improving accuracy while avoiding over-merging
2Reliability
If the expectation is expanded to allow for error, then smaller detections are correctly associated with tracks, but computational complexity increases
Solution Approach 1:
The system applies partial action by selectively expanding the expectation only in spatial dimensions where detection errors are likely to occur, rather than uniformly expanding in all dimensions. This targeted approach maintains association accuracy for smaller detections while limiting the increase in computational complexity to only necessary areas
3Reliability
If motion vectors for individual blocks are calculated to segment objects into clusters, then object fragmentation is reduced, but computational cost increases significantly
Solution Approach 1:
The system extracts only the essential spatial information needed for tracking by deriving an expected spatial representation from previous frames, rather than calculating motion vectors for all individual blocks. This extraction approach maintains object tracking continuity by focusing on the most relevant spatial characteristics while avoiding the computational burden of full block-based motion analysis
Data Source
AI summary
Disclosed herein are a computer-implemented method and a camera system for determining a current spatial representation for a detection in a current frame of an image sequence. The method derives an expected spatial representation (820) for the detection based on at least one previous frame, generates a spatial representation (810) of the detection, and extends the spatial representation (810) to obtain an extended spatial representation (830), based on the expected spatial representation (820). The method determines a similarity measure between the extended spatial representation (830) and the expected spatial representation (820), and then determines the current spatial representation for the detection based on the similarity measure.