The Vision Pre-processing Accelerator (VPAC) provides a set of common vision primitive functions, performing various pixel data processing tasks, such as: color processing and enhancement, noise filtering, wide dynamic range (WDR) processing, lens distortion correction, pixel remap for de-warping, on-the-fly scale generation, on-the-fly pyramid generation, and offloads these common tasks from the main SoC processors (ARM, DSP, etc.). The VPAC includes the following processing and infrastructure sub-modules:
- Vision Imaging Sub-System (VISS) which provides raw data image processing such as:
- Wide Dynamic Range (WDR) merge
- Defect Pixel Correction (DPC)
- Lens Shading Correction (LSC)
- Global/Local Brightness and Contrast Enhancement (GLBCE)
- Advanced Spatial Noise Filter (NSF4V)
- Edge Enhancement (EE)
- Demosaicing
- Color conversion
- Lens Distortion Correction (LDC) block, which provides data reading from memory (DDR or on-chip) and applies perspective transformation as well as correction of lens distortion (including fisheye lenses).
- Multi-Scalar (MSC) block reads data from memory
(DDR or on-chip) to internal shared level 2 (SL2) memory and generates up to
10 scaled outputs from one or two inputs, with various scaling ratios.
- Noise Filter (NF) block reads data from memory (DDR or on-chip) to internal SL2 memory and does Bilateral filtering to remove noise.
- Hardware Thread Scheduler (HTS) provides inter-processor communication among various VPAC sub-modules (VISS, LDC, MSC and NF) as well as with the local DMA Engine (UTC).
- Internal Shared Level 2 (SL2) memory for data exchange across VPAC sub-modules (VISS, LDC, MSC and NF) and from DDR/MSMC, using the K3 Data Movement Architecture (DMA) mechanism
- Load/Store Engine (LSE) which performs data load and store tasks on a the SL2 memory for the hardware accelerator algorithm cores
- VISS to LDC direct OTF (On-the-Fly) for multi-camera
-
YUV422 10 and 12 bit format support on all units except NF (e.g.
LDC/MSC)
-
Simultaneous visual (HV) and analytics (MV) output to system memory
including L3 flex connect, saving need for additional read from system
memory for HV+MV processing.
-
Chromatic Aberration Correction (CAC) to support lower cost lens
- RGBIR support