RadarCFAR

Functions

NVCVStatus pvaRadarCFARCreate(NVCVOperatorHandle *handle, NVCVTensorRequirements const *const inputTensorRequirements, NVCVTensorRequirements const *const outDetectionListTensorRequirements, PVARadarCFARType const algorithm)

Creates and initializes a Radar CFAR operator instance for target detection.

NVCVStatus pvaRadarCFARSubmit(NVCVOperatorHandle handle, cudaStream_t stream, NVCVTensorHandle const inputTensor, NVCVTensorHandle outDetectionListTensor, NVCVTensorHandle outDetectionCountTensor, PvaRadarCFARParams const *const radarCFARParams)

Submits the RadarCFAR operator to a CUDA stream.

NVCVStatus pvaRadarCFARSubmit(NVCVOperatorHandle handle, cupvaStream_t stream, NVCVTensorHandle const inputTensor, NVCVTensorHandle outDetectionListTensor, NVCVTensorHandle outDetectionCountTensor, PvaRadarCFARParams const *const radarCFARParams)

Executes the Radar CFAR target detection operation on input radar data.

Structs

PvaRadarCFARParamsRec

Configuration parameters for radar CFAR target detection operator.

Functions

NVCVStatus pvaRadarCFARCreate(

NVCVOperatorHandle *handle,

NVCVTensorRequirements const *const inputTensorRequirements,

NVCVTensorRequirements const *const outDetectionListTensorRequirements,

PVARadarCFARType const algorithm,

)

Creates and initializes a Radar CFAR operator instance for target detection.

Input Tensor Requirements:

• Data Layout: [HW, HWC] (2D or 3D tensor)

• Channels: [1] (single channel - magnitude data)

• Rank: 2 for [HW] 3 for [HWC] layout

• Width: Valid range (1-1024) and must be divisible by 8

• Height: Valid range (1-1024) and must be divisible by 8

• Height * Width Valid range (1-1024*512)

• Sum of rows or columns: Must not exceed maximum value of 32-bit signed integer

Supported Data Types:

Data Type	Support
8bit Unsigned	❌ Not supported
8bit Signed	❌ Not supported
16bit Unsigned	❌ Not supported
16bit Signed	❌ Not supported
32bit Unsigned	✅ Supported
32bit Signed	✅ Supported
32bit Float	❌ Not supported
64bit Float	❌ Not supported

Detection List Format: The output tensor contains a detection list where each element contains the following fields:

Data Fields
int32_t	rangeIdx
int32_t	dopplerIdx

Output Tensor Characteristics:

• Data Layout: [HW, HWC] (2D or 3D)

• Height: Maximum Detection count. Should be less than 8192

• Width: 2 (number of fields in a detection list entry)

• Channels: [1]

• Rank: 2 for [HW], 3 for [HWC]

Output Data Types:

Data Type	Support
8bit Unsigned	❌ Not supported
8bit Signed	❌ Not supported
16bit Unsigned	❌ Not supported
16bit Signed	❌ Not supported
32bit Unsigned	❌ Not supported
32bit Signed	✅ Supported
32bit Float	❌ Not supported
64bit Float	❌ Not supported

Input/Output Tensor Dependencies:

Property	Input == Output
Data Layout	✅ Must match
Channels	✅ Must match (1 channel)

Note

The operator handle must be destroyed using the appropriate destroy function when no longer needed to prevent memory leaks.

Parameters:

• handle – [out] Pointer to store the created operator handle. Must not be NULL. The handle is used for subsequent operator operations.

• inputTensorRequirements – [in] Pointer to NVCVTensorRequirements structure defining input tensor specifications.

• outDetectionListTensorRequirements – [in] Pointer to NVCVTensorRequirements structure defining output tensor specifications.

• algorithm – [in] PVARadarCFARType containing CFAR algorithm type. Supported algorithms:

  • PVA_CFAR_CA: Cell Averaging

Return values:

• NVCV_ERROR_INVALID_ARGUMENT – Invalid handle pointer or parameter values outside valid range.

• NVCV_ERROR_OUT_OF_MEMORY – Insufficient memory to allocate operator resources.

• NVCV_ERROR_INVALID_IMAGE_FORMAT – Unsupported tensor format or data type.

• NVCV_SUCCESS – Operator created successfully and ready for use.

NVCVStatus pvaRadarCFARSubmit(

NVCVOperatorHandle handle,

cudaStream_t stream,

NVCVTensorHandle const inputTensor,

NVCVTensorHandle outDetectionListTensor,

NVCVTensorHandle outDetectionCountTensor,

PvaRadarCFARParams const *const radarCFARParams,

)

Submits the RadarCFAR operator to a CUDA stream.

Note

CUDA stream support requirements:

• PVA SDK 2.7.0 or later

• Jetpack 7 or later

• DriveOS 7 or later

• x86 Emulator is not supported

Parameters:

• handle – [in] Valid operator handle created by pvaRadarCFARCreate().

• stream – [in] Valid CUDA stream handle for asynchronous execution.

• inputTensor – [in] Input tensor handle containing input data.

• outDetectionListTensor – [out] Output tensor handle for storing detection results.

• outDetectionCountTensor – [out] Output tensor to store the number of detections found.

• radarCFARParams – [in] Pointer to PvaRadarCFARParams structure containing CFAR algorithm configuration.

Return values:

• NVCV_ERROR_INVALID_ARGUMENT – Invalid handle pointer or mismatched tensor dimensions.

• NVCV_ERROR_INTERNAL – Internal processing error, invalid data types, or execution failure.

• NVCV_ERROR_OUT_OF_MEMORY – Insufficient memory for processing.

• NVCV_SUCCESS – CFAR detection completed successfully.

NVCVStatus pvaRadarCFARSubmit(

NVCVOperatorHandle handle,

cupvaStream_t stream,

NVCVTensorHandle const inputTensor,

NVCVTensorHandle outDetectionListTensor,

NVCVTensorHandle outDetectionCountTensor,

PvaRadarCFARParams const *const radarCFARParams,

)

Executes the Radar CFAR target detection operation on input radar data.

This function performs Constant False Alarm Rate (CFAR) target detection on the input radar range-Doppler map data and produces detection list. The operation uses the CFAR algorithm and parameters configured during operator creation to maintain a constant false alarm rate while maximizing target detection probability.

Parameters:

• handle – [in] Valid operator handle created by pvaRadarCFARCreate(). Must not be NULL and must reference a properly initialized operator.

• stream – [in] Valid cuPVA stream handle for asynchronous execution.

• inputTensor – [in] Input tensor handle containing input data.

• outDetectionListTensor – [out] Output tensor handle for storing detection results.

• outDetectionCountTensor – [out] to store the number of detections found.

  • Data type: NVCV_DATA_TYPE_S32

  • Data layout: [W]

  • Length: W = 1;

  • Must not be NULL

  • Contains the count of valid detections in the detection list

• radarCFARParams – [in] Pointer to PvaRadarCFARParams structure containing CFAR algorithm configuration. Must not be NULL. See PvaRadarCFARParams structure documentation for detailed parameter constraints and behavior.

Return values:

• NVCV_ERROR_INVALID_ARGUMENT – Invalid handle pointer or mismatched tensor dimensions.

• NVCV_ERROR_INTERNAL – Internal processing error, invalid data types, or execution failure.

• NVCV_ERROR_OUT_OF_MEMORY – Insufficient memory for processing.

• NVCV_SUCCESS – CFAR detection completed successfully.