Recursive Convolutional Target Detector Efficacy

Folder structure

-----------------

PerformanceEvaluation

  --2peak(Directory)

  --Interference_40%Target_level(Directory)

  --Interference_60%Target_level(Directory)

  --Interference_70%Target_level(Directory)

  --cnnInference.m

  --focusOnQuadrant.m

  --processMatrix.m

  --recursivePeakDetection.m

  --shiftQuadrant.m

  --tracePeakLocation.m

  --PerformanceEval_ROC.m

Description

------------

This research proposes a novel Recursive Convolutional Target Detector (RCTD) for 

Frequency-Modulated Continuous-Wave (FMCW) radar in complex automotive scenarios. 

Leveraging a lightweight convolutional neural network, RCTD efficiently localizes

multiple targets despite strong interference. Detailed simulations and a hardware 

prototype on an FPGA-based deep learning processor demonstrate real-time feasibility,

low false alarm rates, and higher detection accuracy under stringent resource constraints.

The RCTD algorithm is contained in the following Matlab files:

  --cnnInference.m

  --focusOnQuadrant.m

  --processMatrix.m

  --recursivePeakDetection.m

  --shiftQuadrant.m

  --tracePeakLocation.m

The efficacy is evaluted using a Matlab script PerformanceEval_ROC.m. This 

Matlab script loads the Range -Doppler maps data from the different scenarios tested 

and uses classical CFAR and its variants to check performance and similarly checks the 

performance of RCTD algorithm. Finally it plots the results.

The following scenarios were tested:

-- FMCW radar target scenario with single target and interfarence evel of -3 db (This means that the interference power  levels were 3db below the target )

-- FMCW radar target scenario with single target and interfarence evel of 0 db (This means that the interference power levels were same as the target )

-- FMCW radar target scenario with single target and interfarence evel of 1 db (This means that the interference power levels were 1db more that the target power levels )

-- FMCW radar target scenario with 2 targets and interfarence level of -3 db (This means that the interference power levels were 1db more that the target power levels )

The data is contained in the respective folders:

  --2peak                              : 2 targets and interfarence level of -3 db

  --Interference_40%Target_level       : 1 targets and interfarence level of -3 db

  --Interference_60%Target_level       : 1 targets and interfarence level of 0 db

  --Interference_70%Target_level       : 1 targets and interfarence level of 1 db

Run the PerformanceEval_ROC.m to see the results