Neural Network Target Tracking via Local Region Detector
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current single target tracking methods in artificial intelligence face challenges in accurately and efficiently tracking targets across video frames, particularly in automatic driving and multi-target tracking applications, due to variations in target object characteristics and position changes.
Innovation Solution
A target tracking method utilizing a neural network to extract characteristics from template and detection frames, obtaining classification and regression weights, and generating detection boxes through a local region detector to improve tracking accuracy and speed by estimating changes in target position and size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional target tracking methods are used to track targets across video frames, then tracking functionality is provided, but tracking accuracy deteriorates due to variations in target object characteristics and position changes
Solution Approach 1:
The patent employs a neural network to dynamically extract and adapt to changing parameters of the target object across video frames. The network learns to identify target characteristics despite variations in position, scale, and appearance, transforming the tracking system's ability to handle parameter changes rather than relying on fixed thresholds or hand-crafted features.
Solution Approach 2:
The patent replaces traditional mechanical tracking approaches (such as template matching, feature point tracking, or Kalman filtering) with a neural network-based system. This substitution enables the system to automatically learn complex patterns and variations in target appearance, achieving higher accuracy without requiring explicit mathematical models of target motion or appearance changes.
2Measurement precision
If complex target tracking algorithms are employed to improve accuracy, then detection precision is improved, but processing speed deteriorates
Solution Approach 1:
The patent performs preliminary feature extraction and target characteristic learning during the training phase, storing the learned parameters and weights for rapid inference during actual tracking. This preliminary action allows the system to achieve high detection precision during runtime without performing complex computations on every video frame, thereby maintaining processing speed.
3Measurement precision
If traditional detection methods are used to identify target position and size, then simplicity is maintained, but estimation accuracy of target changes deteriorates
Solution Approach 1:
The patent replaces simple geometric or threshold-based detection methods with a neural network that automatically learns to estimate target position, size, and changes. The network substitutes complex manual feature engineering and multiple detection stages with a unified learned model that achieves superior estimation accuracy while maintaining a relatively compact architecture.
Data Source
AI summary
Embodiments of the present disclosure disclose a target detecting method and apparatus, a training method, an electronic device, and a medium. The target detecting method includes: separately extracting, by means of a neural network, characteristics of a template frame and a detection frame, where the template frame is a detection box image of a target object, and the template frame is smaller than the detection frame in image size; obtaining a classification weight and a regression weight of a local region detector based on the characteristic of the template frame; inputting the characteristic of the detection frame into the local region detector to obtain classification results and regression results of multiple alternative boxes output by the local region detector; and obtaining a detection box for the target object in the detection frame according to the classification results and regression results of the multiple alternative boxes output by the local region detector. The embodiments of the present disclosure improve the speed and accuracy in target tracking.