Target Tracking via Position-Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current target tracking methods using deep learning struggle to accurately match and track pedestrians across multiple frame images, especially due to occlusions and position changes, leading to misrecognition and reduced tracking accuracy.
Innovation Solution
A target tracking method that combines non-linear filtering with feature vector analysis to predict the position and identify historical targets based on position and feature differences, using methods like Kalman filters and cosine distance calculations to update and form accurate tracking trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If deep learning method is used to match positions of same person in multiple frame images, then tracking capability is improved, but tracking accuracy deteriorates due to occlusions and position changes
Solution Approach 1:
The patent introduces an intermediary matching mechanism that combines position-based preliminary matching with feature vector verification. The position information serves as an intermediary to narrow down candidate targets, while feature vectors provide the final verification to ensure accurate identification despite occlusions or position changes.
Solution Approach 2:
The patent changes the parameter space by introducing feature vectors as an additional dimension for target identification. Instead of relying solely on position parameters, the system incorporates appearance features, enabling accurate tracking even when position or visibility parameters change due to occlusions.
2Productivity
If position-based matching is used to track targets across frames, then processing speed is improved, but reliability deteriorates due to misrecognition
Solution Approach 1:
The patent segments the target identification process into two independent parts: position-based preliminary matching and feature vector verification. This segmentation allows the system to maintain high processing speed through efficient position-based filtering while ensuring reliability through the subsequent feature verification stage.
Solution Approach 2:
The patent replaces pure mechanical position-based matching with a hybrid approach that incorporates feature vector comparison. This substitution transforms the matching mechanism from a simple spatial calculation to a more robust multi-parameter verification system that reduces misrecognition while maintaining efficiency.
3Measurement precision
If feature vector analysis is added to position-based tracking, then tracking accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by performing position-based matching first to identify candidate targets, thereby reducing the number of feature vector comparisons needed. This preliminary filtering action significantly reduces computational complexity while maintaining high tracking accuracy through the subsequent verification step.
Data Source
AI summary
The present disclosure relates to a target tracking method, device, system, and a non-transitory computer-readable storage medium. The method includes: determining an area where a current target is located by performing target detection on a current frame image; extracting a current position information of the area where the current target is located; predicting a position information of an area where each historical target is located at a corresponding moment of the current frame image based on historical position information of an area where each historical target is located in one or more historical frame images; determining a historical target that is the same target as the current target based on a position difference between the current position information and the predicted position information, and tracking the current target.