Target Tracking Using Position and Appearance Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target tracking methods in automatic drive systems suffer from low accuracy due to reliance on simple information for matching detection boxes across video frames, leading to tracking errors, especially with parallax issues.
Innovation Solution
A target tracking method that utilizes both position information and appearance features, including texture and semantic features, acquired through a neural network model, to match detection boxes across video frames, improving accuracy by calculating a difference degree using adjustable weights for each feature distance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple information is used for matching detection boxes, then the matching process is fast and easy, but tracking accuracy is low
Solution Approach 1:
The patent combines multiple feature types (position information, texture features, and semantic features) into a unified matching framework. The appearance feature extraction module extracts both texture and semantic features from detection box images, and the matching module integrates all feature types to determine matching relationships, thereby improving tracking accuracy without excessive complexity increase
Solution Approach 2:
The patent transitions from using only simple position information to incorporating multi-dimensional feature spaces including texture features (low-level visual patterns) and semantic features (high-level object understanding). This dimensional expansion allows the system to distinguish between objects more accurately, especially in cases of overlap and parallax
2Reliability
If only position information is used for matching, then the system is simple and fast, but tracking errors occur with parallax problems
Solution Approach 1:
The patent introduces appearance features (texture and semantic) as intermediary elements between position information and final matching decisions. These appearance features serve as mediators that help disambiguate parallax issues by providing additional visual context about object characteristics, enabling more reliable tracking while maintaining system manageability
Solution Approach 2:
The patent creates a composite feature representation that integrates multiple feature types (position, texture, semantic) into a unified matching criterion. This composite approach leverages the strengths of each feature type while compensating for their individual limitations, improving reliability especially in challenging parallax scenarios
3Measurement precision
If multiple feature types are used for matching, then tracking accuracy is improved, but computational time increases
Solution Approach 1:
The patent segments the feature extraction process into distinct modules: position information extraction, texture feature extraction, and semantic feature extraction. Each module processes specific aspects of the detection box independently, allowing for optimized computation and enabling parallel processing to reduce overall computational time while maintaining high matching accuracy
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A target tracking method and device, as well as an electronic apparatus and a storage medium are provided according to embodiments of the present disclosure. The method includes: acquiring feature information of a detection box in a current video frame and feature information of a detection box in a previous video frame, the feature information comprising position information of the detection box and an appearance feature of an obstacle target within the detection box, wherein the appearance feature comprises a texture feature and a semantics feature; matching the feature information of the detection box in the current video frame with the feature information of the detection box in the previous video frame; and determining a tracking trajectory of the detection box in the current video frame according to a matching result. In the embodiment of the application, an accuracy for tracking an object can be improved.