sources/includes/nvll_obb.h File Reference

Go to the source code of this file.

Typedefs
typedef struct _NvOBBDumpCtx	NvOBBDumpCtx
	OBB dump context handle. More...

Functions
NvOBBDumpCtx *	nvobb_dump_create_context (int gpu_id, cudaStream_t cuda_stream)
	Create OBB dump context. More...
void	nvobb_dump_destroy_context (NvOBBDumpCtx *ctx)
	Destroy OBB dump context. More...
NvOBBDumpCtx *	nvobb_dump_get_singleton (int gpu_id, cudaStream_t cuda_stream)
	Get or create singleton OBB dump context. More...
void	nvobb_dump_cleanup_singleton (void)
	Cleanup singleton OBB dump context. More...
bool	nvobb_dump_frame (NvOBBDumpCtx *ctx, NvBufSurface *input_surf, void *rect_params_array, int num_rects, unsigned int frame_num)
	Process frame and dump OBBs. More...
cudaError_t	nvobb_rotate_gpu (const uint8_t *aabb_pixels, int aabb_width, int aabb_height, int aabb_pitch, int obb_width, int obb_height, float obb_cx, float obb_cy, float angle_rad, cudaStream_t stream, uint8_t **d_output)
	Rotate OBB to rectangular RGBA on GPU (standalone) More...
uint8_t *	nvobb_rotate_cpu (const uint8_t *aabb_pixels, int aabb_width, int aabb_height, int aabb_pitch, int obb_width, int obb_height, float obb_cx, float obb_cy, float angle_rad)
	Rotate OBB to rectangular RGBA on CPU (standalone) More...

Typedef Documentation

NvOBBDumpCtx

typedef struct _NvOBBDumpCtx NvOBBDumpCtx

OBB dump context handle.

Note: NvOSD_RectParams and NvOSD_ColorParams are defined in nvll_osd_struct.h and are included via nvll_osd_api.h

Definition at line 30 of file sources/includes/nvll_obb.h.

Function Documentation

nvobb_dump_cleanup_singleton()

void nvobb_dump_cleanup_singleton

(

void

)

Cleanup singleton OBB dump context.

Should be called during application shutdown to properly release all OBB dump resources. Thread-safe and idempotent.

nvobb_dump_create_context()

NvOBBDumpCtx* nvobb_dump_create_context	(	int	gpu_id,
		cudaStream_t	cuda_stream
	)

Create OBB dump context.

Initializes the OBB dumping system. Reads configuration from environment variables:

• NVDS_OBB_DUMP_ENABLED: Set to 1 to enable (default: disabled)
• NVDS_OBB_DUMP_DIR: Output directory (default: /tmp/obb_dumps)
• NVDS_OBB_DUMP_QUALITY: JPEG quality 1-100 (default: 90)
• NVDS_OBB_DUMP_FRAME_INTERVAL: Dump every Nth frame (default: 30)
• NVDS_OBB_DUMP_MIN_ANGLE: Minimum angle threshold in degrees (default: 1.0)

Parameters

gpu_id	GPU device ID
cuda_stream	CUDA stream to use for async operations

Returns

NvOBBDumpCtx* Context handle, or NULL if disabled or failed

nvobb_dump_destroy_context()

void nvobb_dump_destroy_context

(

NvOBBDumpCtx *

ctx

)

Destroy OBB dump context.

Waits for all pending encodes to finish and frees all resources.

Parameters

ctx	Context handle

nvobb_dump_frame()

bool nvobb_dump_frame	(	NvOBBDumpCtx *	ctx,
		NvBufSurface *	input_surf,
		void *	rect_params_array,
		int	num_rects,
		unsigned int	frame_num
	)

Process frame and dump OBBs.

Extracts and saves rotated OBBs from the frame. This function:

1. Checks frame interval (only processes every Nth frame)
2. For each OBB with angle >= min_angle: a. Calculates AABB containing the rotated OBB b. Crops AABB using NvBufSurfTransform (GPU) c. Extracts rotated pixels using obb_scanline_rotate_gpu d. Encodes to JPEG using nvds_obj_encode

Parameters

ctx	Context handle
input_surf	Input surface containing the frame
rect_params_array	Array of rectangle parameters (NvOSD_RectParams*)
num_rects	Number of rectangles in the array
frame_num	Frame number (for filename generation)

Returns

bool true on success, false on failure

nvobb_dump_get_singleton()

NvOBBDumpCtx* nvobb_dump_get_singleton	(	int	gpu_id,
		cudaStream_t	cuda_stream
	)

Get or create singleton OBB dump context.

Thread-safe lazy initialization of global OBB dump context. The context is created once and reused across all OSD drawing calls.

Parameters

gpu_id	GPU device ID
cuda_stream	CUDA stream (0 for default stream)

Returns

Pointer to singleton context, or NULL if disabled/failed

nvobb_rotate_cpu()

uint8_t* nvobb_rotate_cpu	(	const uint8_t *	aabb_pixels,
		int	aabb_width,
		int	aabb_height,
		int	aabb_pitch,
		int	obb_width,
		int	obb_height,
		float	obb_cx,
		float	obb_cy,
		float	angle_rad
	)

Rotate OBB to rectangular RGBA on CPU (standalone)

This function performs ONLY the rotation on CPU. No dumping, no threading, no encoding. Use this when you want to process rotated OBBs yourself and don't need GPU acceleration.

Parameters

aabb_pixels	Input AABB pixels (CPU memory, RGBA format)
aabb_width	Width of input AABB
aabb_height	Height of input AABB
aabb_pitch	Pitch (stride) of AABB in bytes
obb_width	Width of output OBB
obb_height	Height of output OBB
obb_cx	Center X of OBB in AABB space
obb_cy	Center Y of OBB in AABB space
angle_rad	Rotation angle in radians

Returns

uint8_t* Allocated CPU buffer with rotated OBB (RGBA), or NULL on failure Caller must free with g_free() or free()

Note

Memory management: Caller is responsible for freeing returned buffer with g_free()

Output format: RGBA, tightly packed, row-major, size = obb_width × obb_height × 4 bytes

nvobb_rotate_gpu()

cudaError_t nvobb_rotate_gpu	(	const uint8_t *	aabb_pixels,
		int	aabb_width,
		int	aabb_height,
		int	aabb_pitch,
		int	obb_width,
		int	obb_height,
		float	obb_cx,
		float	obb_cy,
		float	angle_rad,
		cudaStream_t	stream,
		uint8_t **	d_output
	)

Rotate OBB to rectangular RGBA on GPU (standalone)

This function performs ONLY the rotation on GPU. No dumping, no threading, no encoding. Use this when you want to process rotated OBBs yourself (e.g., for tracking, feature extraction, etc.)

Parameters

	aabb_pixels	Input AABB pixels (GPU memory, RGBA format)
	aabb_width	Width of input AABB
	aabb_height	Height of input AABB
	aabb_pitch	Pitch (stride) of AABB in bytes
	obb_width	Width of output OBB
	obb_height	Height of output OBB
	obb_cx	Center X of OBB in AABB space
	obb_cy	Center Y of OBB in AABB space
	angle_rad	Rotation angle in radians
	stream	CUDA stream for async execution (use 0 for default stream)
[out]	d_output	Pointer to receive allocated GPU buffer with rotated OBB Caller must free with cudaFree()

Returns

cudaError_t cudaSuccess on success, error code on failure

Note

Memory management: Caller is responsible for freeing *d_output with cudaFree()

Output format: RGBA, tightly packed, row-major, size = obb_width × obb_height × 4 bytes