Video Object Association via Spatial and Appearance Distance Measures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection systems in videos suffer from object identity switching and loss when objects are close together, due to the lack of consideration for temporal relations between frames, leading to inaccurate tracking and stability issues.
Innovation Solution
A computer-implemented method that associates objects in subsequent video frames by using distance measures calculated from appearance measures, spatial locations, and detection probabilities, allowing for improved tracking and stability of object identities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional object detectors are applied individually on each frame, then processing speed is maintained, but object identity switching occurs when objects are close together
Solution Approach 1:
The system performs preliminary actions by obtaining object proposal regions and their attributes (spatial location, appearance measure, detection probability) from the previous frame before processing the current frame. This temporal preparation enables the association algorithm to correctly identify and track objects across frames, preventing identity switching while maintaining real-time processing capability.
Solution Approach 2:
The patent introduces an intermediary association algorithm that acts as a mediator between consecutive frame detections. This algorithm uses distance measures calculated from spatial locations, appearance measures, and detection probabilities to correctly associate objects across frames, resolving the identity switching problem without requiring complex temporal processing of entire video sequences.
2Measurement precision
If detection probability threshold is applied strictly, then false positives are reduced, but objects are lost when detection probability temporarily drops below threshold
Solution Approach 1:
The system implements feedback by utilizing detection probabilities from the previous frame as input for the current frame's object association. When an object's detection probability temporarily drops below the threshold, the association algorithm uses historical detection probability information and appearance measures to maintain tracking continuity, preventing object loss while preserving detection accuracy through probabilistic reasoning.
3Reliability
If smoothing detection probabilities is applied, then object tracking continuity is improved, but accuracy is reduced due to low precision in probability smoothing
Solution Approach 1:
Instead of smoothing detection probabilities which degrades accuracy, the patent changes the approach by using distance measures calculated from multiple parameters including spatial location, appearance measure, and detection probability. This multi-parameter association method maintains both tracking continuity and detection accuracy by avoiding probability smoothing while still leveraging temporal information from previous frames.
Data Source
AI summary
The invention relates to a computer implemented method for associating objects in a video comprising subsequent frames, the method comprising obtaining first object proposal region information of a previous frame, determining second object proposal region information of a current frame, wherein the first and second object proposal region information are at least indicative of an appearance measure, a spatial location and a detection probability of each object proposal region of the respective frame, associating objects in the video by at least associating a first set of object proposal regions of the previous frame to a second set of object proposal regions of the current frame, wherein the object proposal regions are associated using distance measures calculated based on the appearance measures, the spatial locations and the detection probabilities.