Harris Corner#

Compute the Harris response for each input pixel in a tile.

This algorithm follows these steps:

1) Apply 3x3 gradient operator filters to determine each of the x and y gradients. The coefficients of the filters are given by [ coeff ]^T * [-1 0 1] for x gradients and [-1 0 1]^T * [ coeff ] for y gradients.

2) Construct M matrix for each pixel. M is computed by summing gradient covariance matrices in a 3x3 block for each pixel.

3) Set each output pixel to R = det(M) - lambda * trace^2(M)

The combination of the 3x3 gradient operator filter with 3x3 block summation means that each output pixel has 5x5 support in the input. The valid output will be 4 pixels less than the input in each width and height dimensions and offset by 2 - i.e. the output at x=0, y=0 is the Harris response of input pixel x=2, y=2.

Macros#

PVAAPL_HARRIS_SCRATCH_SIZE

Determine bytes needed for each scratch buffer in Harris corner detector.

Functions#

void pvaAplExecHarrisCornerS16(PvaAplHarrisCornerS16 *handle)

Execute the Harris Corner algorithm with a given handle.

void pvaAplExecHarrisCornerS16Vpu(PvaAplHarrisCornerS16 *handle)

Execute the Harris Corner algorithm with a given handle.

void pvaAplInitHarrisCornerS16(PvaAplHarrisCornerS16 *handle, int16_t *input, int16_t *coeff, int32_t roundNBits, int32_t lambda, int32_t width, int32_t height, int32_t inputLinePitch, int32_t outputLinePitch, int32_t *output, int16_t *cbStart, int32_t cbSize, void *scratch0, void *scratch1, void *scratch2)

Initialize the Harris Corner Score algorithm parameters.

void pvaAplInitHarrisCornerS16Vpu(PvaAplHarrisCornerS16 *handle, int16_t *input, int16_t *coeff, int32_t roundNBits, int32_t lambda, int32_t width, int32_t height, int32_t inputLinePitch, int32_t outputLinePitch, int32_t *output, int16_t *cbStart, int32_t cbSize, void *scratch0, void *scratch1, void *scratch2)

Initialize the Harris Corner Score algorithm parameters.

void pvaAplUpdateHarrisCornerS16(PvaAplHarrisCornerS16 *handle, int16_t *input, int32_t *output)

Update the input/output buffer pointers and input image size for the Harris Corner Score algorithm.

void pvaAplUpdateHarrisCornerS16Vpu(PvaAplHarrisCornerS16 *handle, int16_t *input, int32_t *output)

Update the input/output buffer pointers and input image size for the Harris Corner Score algorithm.

Functions#

inline void pvaAplExecHarrisCornerS16(PvaAplHarrisCornerS16 *handle)#

Execute the Harris Corner algorithm with a given handle.

Parameters:

handle – Pointer to the Harris Corner algorithm handle.

void pvaAplExecHarrisCornerS16Vpu(PvaAplHarrisCornerS16 *handle)#

Execute the Harris Corner algorithm with a given handle.

This variant ensures synchronous execution on VPU, even on platforms supporting asynchronous acceleration. Do not call pvaAplWait() when using this variant.

Parameters:

handle – Pointer to the Harris Corner algorithm handle.

inline void pvaAplInitHarrisCornerS16(
PvaAplHarrisCornerS16 *handle,
int16_t *input,
int16_t *coeff,
int32_t roundNBits,
int32_t lambda,
int32_t width,
int32_t height,
int32_t inputLinePitch,
int32_t outputLinePitch,
int32_t *output,
int16_t *cbStart,
int32_t cbSize,
void *scratch0,
void *scratch1,
void *scratch2,
)#

Initialize the Harris Corner Score algorithm parameters.

This algorithm makes use of scratch buffers if executed on VPU. Each scratch buffer should be aligned to 64B and with size given by PVAAPL_HARRIS_SCRATCH_SIZE. For best performance, place all three scratch buffers in separate superbanks. Scratch buffers can be NULL if VPU is not used.

Parameters:

  • handle – Pointer to the algorithm handle.

  • input – Pointer to the buffer in int16_t pixel format. The recommended input pixel value range is [0, 255] or [-128, 127]. Output may saturate if the input pixel value range is broader.

  • coeff – Coefficients for the averaging part of the gradient operator. These coefficients are combined with a differentiation filter to create the final 3x3 gradient operators. Horizontal Gradient Operator: Gx = [coeff[0], coeff[1], coeff[2]]^T * [-1 0 1] Vertical Gradient Operator: Gy = [-1 0 1]^T * [coeff[0], coeff[1], coeff[2]] Should point to 3 element array of int16_t format.

  • roundNBits – Right shift amount for rounding the output harris corner score. Harris corner score is internally computed in 48-bit registers truncated to 32-bit before storing the result. Recommended value is 5. Can be increased depending on the input data, gradient operator coefficients and the lambda parameter.

  • lambda – Constant value used in the Harris corner score computation. R = det(M) - lambda * trace^2(M) Should be passed as fixed point 16-bit quantized value. For example to pass 0.05: (int32_t)(0.05f * (float)(1 << 16)) Recommended range is 0.04 to 0.06.

  • width – Input image width in pixels.

  • height – Input image height in pixels.

  • inputLinePitch – Input buffer line pitch in pixels.

  • outputLinePitch – Output buffer line pitch in pixels. Must be at least (width - 4)

  • output – Pointer to the output Harris corner score buffer in int32_t format. Output is written in blocks of 16 rows. The buffer must be large enough to hold height - 4 rows, rounded up to a multiple of 16. For example if input height is 136, the buffer should be large enough to hold 144 rows. Only the first height - 4 rows will contain valid data.

  • cbStart – Start address of the input circular buffer. Pass NULL if circular buffer is not used.

  • cbSize – Size of the input circular buffer in bytes. Pass 0 if circular buffer is not used.

  • scratch0 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

  • scratch1 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

  • scratch2 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

void pvaAplInitHarrisCornerS16Vpu(
PvaAplHarrisCornerS16 *handle,
int16_t *input,
int16_t *coeff,
int32_t roundNBits,
int32_t lambda,
int32_t width,
int32_t height,
int32_t inputLinePitch,
int32_t outputLinePitch,
int32_t *output,
int16_t *cbStart,
int32_t cbSize,
void *scratch0,
void *scratch1,
void *scratch2,
)#

Initialize the Harris Corner Score algorithm parameters.

This algorithm makes use of scratch buffers if executed on VPU. Each scratch buffer should be aligned to 64B and with size given by PVAAPL_HARRIS_SCRATCH_SIZE. For best performance, place all three scratch buffers in separate superbanks. Scratch buffers can be NULL if VPU is not used.

This variant ensures synchronous execution on VPU, even on platforms supporting asynchronous acceleration.

Parameters:

  • handle – Pointer to the algorithm handle.

  • input – Pointer to the buffer in int16_t pixel format. The recommended input pixel value range is [0, 255] or [-128, 127]. Output may saturate if the input pixel value range is broader.

  • coeff – Coefficients for the averaging part of the gradient operator. These coefficients are combined with a differentiation filter to create the final 3x3 gradient operators. Horizontal Gradient Operator: Gx = [coeff[0], coeff[1], coeff[2]]^T * [-1 0 1] Vertical Gradient Operator: Gy = [-1 0 1]^T * [coeff[0], coeff[1], coeff[2]] Should point to 3 element array of int16_t format.

  • roundNBits – Right shift amount for rounding the output harris corner score. Harris corner score is internally computed in 48-bit registers truncated to 32-bit before storing the result. Recommended value is 5. Can be increased depending on the input data, gradient operator coefficients and the lambda parameter.

  • lambda – Constant value used in the Harris corner score computation. R = det(M) - lambda * trace^2(M) Should be passed as fixed point 16-bit quantized value. For example to pass 0.05: (int32_t)(0.05f * (float)(1 << 16)) Recommended range is 0.04 to 0.06.

  • width – Input image width in pixels.

  • height – Input image height in pixels.

  • inputLinePitch – Input buffer line pitch in pixels.

  • outputLinePitch – Output buffer line pitch in pixels. Must be at least (width - 4)

  • output – Pointer to the output Harris corner score buffer in int32_t format. Output is written in blocks of 16 rows. The buffer must be large enough to hold height - 4 rows, rounded up to a multiple of 16. For example if input height is 136, the buffer should be large enough to hold 144 rows. Only the first height - 4 rows will contain valid data.

  • cbStart – Start address of the input circular buffer. Pass NULL if circular buffer is not used.

  • cbSize – Size of the input circular buffer in bytes. Pass 0 if circular buffer is not used.

  • scratch0 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

  • scratch1 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

  • scratch2 – Scratch buffer. Size needed is given by PVAAPL_HARRIS_SCRATCH_SIZE.

inline void pvaAplUpdateHarrisCornerS16(
PvaAplHarrisCornerS16 *handle,
int16_t *input,
int32_t *output,
)#

Update the input/output buffer pointers and input image size for the Harris Corner Score algorithm.

Parameters:

  • handle – Pointer to the algorithm handle.

  • input – Pointer to the input image

  • output – Pointer to the output Harris corner score buffer.

void pvaAplUpdateHarrisCornerS16Vpu(
PvaAplHarrisCornerS16 *handle,
int16_t *input,
int32_t *output,
)#

Update the input/output buffer pointers and input image size for the Harris Corner Score algorithm.

This variant ensures synchronous execution on VPU, even on platforms supporting asynchronous acceleration.

Parameters:

  • handle – Pointer to the algorithm handle.

  • input – Pointer to the input image

  • output – Pointer to the output Harris corner score buffer.