Radar Target Tracking Using Recurrent Neural Network Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional radar target tracking techniques face challenges in accurately tracking targets due to noise and changes in dynamics, velocity, and acceleration, leading to increased computational complexity and failure probabilities.
Innovation Solution
The implementation of a recurrent neural network with a memory function, trained with a maximum of Q tracking buffers, is used to identify target signals amidst noise and fake signals, combined with classical tracking algorithms to enhance accuracy and stability, and the ability to reprocess outputs and adjust processing windows for improved tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional tracking algorithms (Kalman filter, PDAF, JPDAF) are used to estimate future paths based on current and past paths, then the tracking system can provide prediction capability, but noise makes accurate measurement difficult and the probability of failure increases when dynamics, velocity, or acceleration change
Solution Approach 1:
The patent segments the tracking problem into multiple independent neural network processors, each handling a specific target. Each processor independently processes radar signals and maintains tracking buffers, avoiding the computational complexity of centralized probabilistic algorithms while maintaining reliability through distributed processing
Solution Approach 2:
The patent replaces conventional mechanical/mathematical filtering algorithms (Kalman filter, PDAF, JPDAF) with an artificial neural network-based system. The neural network learns target patterns and dynamics from training data, substituting complex probabilistic calculations with learned representations that are more robust to noise and dynamic changes
2Measurement precision
If the processing window is set small to focus on the target, then tracking precision is improved, but when tracking fails the system cannot recover without enlarging the window and retrying
Solution Approach 1:
The patent implements a dynamic processing window that automatically adjusts its size and position based on tracking confidence and target dynamics. When tracking is confident, the window remains small for precision; when uncertainty increases or tracking fails, the window automatically enlarges to capture more context and enable recovery, then refines back to precise tracking
3Productivity
If multiple targets are tracked using parallel processing, then the system can handle multiple objects simultaneously, but the computational load increases with the number of targets
Solution Approach 1:
The patent segments the multi-target tracking problem into independent parallel processors, each dedicated to a specific target. This segmentation allows simultaneous tracking of multiple targets while optimizing computational resources for each individual track, reducing overall energy consumption compared to a monolithic processing approach
Solution Approach 2:
Each neural network processor is self-contained and independently processes its assigned target using its own tracking buffers and decision logic. This self-service architecture eliminates the need for complex inter-processor communication and coordination, reducing computational overhead and energy consumption while maintaining high productivity in multi-target scenarios
Data Source
AI summary
Disclosed is a technique for processing signals received from a radar and, in particular, a technique for tracking a target on the basis of detected target candidate signals. The proposed invention introduces a recurrent neural network with a memory function in order to find a target signal from signals with noise and fake signals mixed therein. This recurrent neural network is trained to have a maximum of Q tracking buffers therein. According to an additional aspect, it is possible to increase tracking accuracy through a serial connection of the recurrent neural network. According to an additional aspect, it is possible to track multiple targets through a parallel connection of the recurrent neural network.