Target Tracking With Optical Flow and Kalman Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target tracking algorithms, such as SORT, suffer from tracking ID switching and missing detection due to reliance on simple Intersection Over Union (IOU) matching, leading to unstable tracking and high resource consumption.
Innovation Solution
A combined method using target detection, optical flow calculation, Kalman Filter prediction, and multiple matching stages to stabilize tracking, including optical flow point resampling and Kalman motion model updates, to enhance tracking stability and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If SORT algorithm uses only IOU matching between adjacent frames, then the algorithm has low complexity and fast running speed, but it causes tracking ID switching and missing detection leading to unstable tracking
Solution Approach 1:
The patent segments the matching process into multiple stages: first matching between adjacent frames using IOU, and second matching between non-adjacent frames using enhanced IOU calculation. This segmentation allows the system to maintain fast initial matching while adding a stabilization layer that corrects tracking errors without significantly increasing overall complexity.
Solution Approach 2:
The patent performs preliminary tracking using standard IOU matching to establish initial trajectories, then applies preliminary correction actions through the second matching stage that identifies and corrects potential ID switching errors before they propagate. This preliminary action approach maintains speed while preventing reliability issues.
2Device complexity
If SORT algorithm relies heavily on detection results, then the algorithm structure remains simple, but it is easy to cause jump of tracking ID number due to missing detection
Solution Approach 1:
The patent introduces an intermediary verification mechanism that acts as a mediator between detection results and tracking decisions. The second matching stage serves as an intermediary layer that validates detection results and corrects errors without requiring complete redesign of the detection-tracking pipeline, thus maintaining structural simplicity while improving reliability.
Solution Approach 2:
The patent implements feedback mechanisms where the second matching stage provides corrective feedback to the tracking system when detection errors are detected. This feedback loop allows the system to self-correct tracking ID jumps caused by missing detections without adding complex structural changes to the overall algorithm.
3Device complexity
If matching only relies on IOU between adjacent frames, then the calculation is simple, but it easily leads to exchange of trace object IDs
Solution Approach 1:
The patent extends the matching process from a single dimension (adjacent frames only) to multiple dimensions by adding a second matching stage that considers non-adjacent frames. This dimensional extension allows the system to maintain simple IOU calculations while improving identification accuracy through multi-frame context analysis.
Solution Approach 2:
The patent creates a composite matching approach that combines standard IOU matching with enhanced IOU verification. This composite strategy integrates two matching methods with different strengths, using the simple IOU for speed and the enhanced IOU for accuracy, thereby improving object identification without significantly increasing overall complexity.
Data Source
AI summary
Disclosed are a target tracking method, a target tracking system and an electronic device. The target tracking method includes acquiring N frames of images sequentially, wherein N is an integer and N≥2. The target tracking method further includes: performing target detection on an N-th frame of image to generate a target area; performing optical flow calculation and Kalman Filter prediction on a target area in an (N−1)-th frame of image to generate an optical flow predicted target area and a Kalman Filter predicted target area in the N-th frame of image, respectively; performing a first matching between the target area in the N-th frame of image and the optical flow predicted target area in the N-th frame of image; performing second matching on an unmatched target area in an N-th frame of image and the Kalman Filter predicted target area in the N-th frame of image.