- Image Size: Configurable.
- Input Format: 32/16 - bit PMA* output.
- Output Format:
- Number of detected objects
- Properties of detected objects, such as start co-ordinates, size, margin etc.
- Fully re-entrant and multi-instancing capable
- Code compatible across the ADSP-BF60x and ADSP-BF70x Blackfin Processor Family
- Framework Dependencies: None
- Release Format: Object code module with C source wrapper.
* Polar Magnitude and Angle (PMA): is a block in PVP available in ADSP-BF609. If the bits per sample is 32, the input should be the port 2 output of the PMA block of the PVP. The lower 16 bits contain the gradient magnitude and the upper 16 bits contain the gradient angle. If the bits per sample is 16, the lower 11 bits contain the gradient magnitude and upper 5 bits contain the gradient angle information. Please refer to Hardware Reference Manual of ADSP-BF609 for more detail.
The Blackfin HOG SVM Detector is a Software Module meant for computer vision applications targeting Blackfin family of processors. The Blackfin HOG SVM Detector comes with an optimized implementation of Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM) classifier. This Software Module can be used for detection of various objects such as people, pets, cars etc in a variety of applications.
The Blackfin HOG SVM Detector module from ADI has been highly optimized to run on the Analog Devices’ Blackfin processor family. It is a self-contained software module based on an ADI proprietary design and has been tested with a variety of images. The Blackfin HOG SVM Detector module is fully re-entrant and multi-instance capable, and comes with an intuitive API which makes it easy for system integration.
Each module supports the Analog Devices, Inc. (ADI) Blackfin or SHARC Processor family and is a licensed product that is available in object code format. Recipients must sign or accept a license agreement with ADI prior to being shipped or downloading the modules identified in the license agreement.
|Processor||Code||Stack||Data RAM||Data ROM||Average Cycles / pel||MIPS|
||36.54 (HOG SVM Detector),
26.88 (1-D Gradient and PMA)
|378 (HOG SVM Detector),
278 (1-D Gradient and PMA)
- Performance is measured on the following processors using ADI CrossCore® Embedded Studio 2.0.0 tool-chain
- ADSP-BF609 : CCLK = 500 MHz, SCLK = 250 MHz
- ADSP-BF707 : CCLK = 384 MHz, SCLK = 192 MHz
- Instruction cache and MDMA enabled. Data Cache is enabled on Bank A.
- MIPS is calculated for a video of 720x480 resolution @ 30 fps
- On ADSP-BF609 1-D Gradient and PMA computation is performed using PVP
- “Data RAM” for one instance includes, scratch and state memory but excludes input/output buffers
- Performance figures depend on configuration parameters and input vector attributes. The figures given in the table are for typical use case with following parameters.
- Number of Classifiers: 1
- Window Size: (64x128) with only Primary classifier
- Window Stride: 1
- 1 KiB = 1024 Bytes
- Windows 7 Professional/Enterprise/Ultimate (32 and 64-bit).
- Minimum of 2 GHz single core processor, 3.3 GHz dual core is recommended.
- Minimum of 1 GB memory (RAM), 4 GB is recommended.
- Minimum of 2 GB hard disk (HDD) space is required.
- CrossCore Embedded Studio for Analog Devices Processors.