topic Re: AN4629 Source in i.MX Processors

AN4629 Source

JohnCannizzo — Sat, 06 Jun 2015 06:49:42 GMT

I found App Note AN4629 very interesting. I'm trying to port surveillance and face recognition from the TI DM368 "F" to the iMX6. I'm fairly familiar with both platforms, but this App Note was excellent in that it showed how to use the VPU for object detection / recognition and provided references for further information. I didn't see the authors name in the App Note, but I wanted to offer my thanks for a well written and research App Note.

I was wondering if the source code for the App Note was available? I'd love to see a complete application based on this VPU approach, and tweak it to observe frame rates for face recognition at various resolutions.

Thanks,

John

Re: AN4629 Source

igorpadykov — Tue, 09 Jun 2015 00:19:01 GMT

Hi John

I believe one can use document on link below

Step 2 – Building and installing OpenCV

Building OpenCV on ubuntu hardfloat rootfs for i.MX6Q

Best regards

igor

-----------------------------------------------------------------------------------------------------------------------

Note: If this post answers your question, please click the Correct Answer button. Thank you!

-----------------------------------------------------------------------------------------------------------------------

Re: AN4629 Source

JohnCannizzo — Wed, 10 Jun 2015 06:33:18 GMT

Hi Igor,

I'm ordering parts to enable MIPI on my Sabre board, so I haven't tried this yet, but in reading through the document, the sample appears to perform face detection on a still JPEG image using OpenCV. The App Note I referenced uses the VPU to perform face detection and recognition on a video stream using the VPU, since OpenCV is too slow to run face recognition anywhere near 30 fps on even a 320 x 240 video stream (according to the App Note). I'm working on a surveillance camera with embedded analytics, so there is no attached display. Offloading as many of the analytics from the main processor(s) onto the VPU is my goal, and is the topic of the App Note. The code used in the App Note offers a huge head start in my efforts, and would allow me to implement new analytics into it to test frame rates independent of the surveillance application.

If the code is available, I'd love to take a look...

Thanks again,

John

Re: AN4629 Source

igorpadykov — Wed, 10 Jun 2015 15:01:05 GMT

Hi John

could you provide full name of document (with web link) and page,

where it describes that "VPU to perform face detection" ?

Best regards

igor

Re: AN4629 Source

JohnCannizzo — Thu, 11 Jun 2015 21:06:39 GMT

Hi Igor,

The link to the App Note is http://cache.freescale.com/files/32bit/doc/app_note/AN4629.pdf

The document title is

Fast Image Processing with i.MX 6 Series

Here is an sample from the text:

The goal of this application note is to present the reader with concrete techniques, mathematical background, and code examples that will allow them to perform Image Processing using the Freescale i.MX 6 series reference boards. We will be focusing on:

1. Image acquisition

2. Segmentation techniques

3. Image filter operators

4. Morphologic operations

5. Image binarization

6. Tracking techniques

7. Feature extraction and its uses

8. Face detection techniques

As stated before, some of these algorithms will be implemented in traditional C++ code that will run on the CPU. However, the goal of this application note is to show the reader how to implement several OpenGL ES 2.0 shaders that will run on the GPU and will do the most intense and operation-heavy work in order to accelerate the image processing tasks.

I must admit that I was using VPU and GPU interchangeably, which is a bad habit on ARM processors. I meant the GPU of the iMX6, which is the topic of the App Note.

Thanks,

John

Re: AN4629 Source

igorpadykov — Fri, 12 Jun 2015 11:15:06 GMT

Hi John

now I agree with you, appnote uses and describes GPU, not VPU.

My previous remark was just answer on your update:

"The App Note I referenced uses the VPU to perform face detection.."

Best regards

igor

Re: AN4629 Source

JohnCannizzo — Thu, 18 Jun 2015 20:06:16 GMT

Any update here? I'm still looking for the code used in the App Note. Thanks...