The face recognition system is implemented on the Ultra96-V2 development board, which uses the VITRA96-V2 development board to better create a face application, in which the following models use the following model:
WorkDetect: Face detection (DensityBox_640_360)
Worktrack: Face Tracking
Workcapture: Facial-Quality
Workrecog: Human face characteristics (FaceRec_resnet 20)
WorkCompare: Find the matching face
Hardware components:
ULTRA96-V2
Camera
DisplayPort Monitor
Design steps:
Step 1, create an SD card
Provide pre-built VITIS-AI 1.3 SD card images for the AVNET platform:
U96V2_SBC_BASE: ULTRA96-V2 development board
UZ7EV_EVCC_BASE: Ultrazed-EV SOM (7EV) + FMC operator card
UZ3EG_IOCC_BASE: Ultrazed-Eg Som (3EG) + IO Operator Card
After downloading and decompression, you can program the .img file to 16GB Micro SD card:
1, unzip file, get .img file
2, program the SD card image of the motherboard to 16GB (or more) Micro SD card. On Windows Computers, use Balena etcher or Win32Diskimager (free open source software)
3. On Linux machines, use Balena etcher or use DD utilities:
$ sudo dd bs = 4m if = avnet- {platform} -VITIS-AI-1-3- {DATE} .img of = / dev / sd {x} status = progress conv = fsync
(Where {x} is lowercase letters, used to specify the SD card device. You can use "DF -H" to determine which device corresponds to your SD card)
Step 2, download the face recognition code
1. Register Xilinx (click to enter)
2. Click the "Demo and Design" tab, you can find some sources of application examples (including facial recognition)
Download "FACE_RECOG_DEMO_SRC_V1.3.TAR.GZ" file and copy it to the SD card of the Ultra96-V2 board
Step three, extract and build a face recognition
"FACE_RECOG_DEMO_SRC_V1.3.TAR.GZ" archive contains many useless files (. *), You can use the --exclude parameter to extract files to prevent some unwanted files
After extracting the archive, you can build an application. Since the application's X-Windows version is to be used, it is built with build_x.sh scripts.
The first attempt to build an application on the Ultra96-V2 caused the memory inadequate error, because the call is caused by the "-j" parameter (ie 10). To resolve this issue, modify the cmake.sh script and change the following code:
You can use the SED to operate:
Construction success:
Create two executables:
Build_Database: Database for creating facial features
Face_Demo: Used to identify people in the video
Step 4, execute the image / video
Create a face database: Create a facial database file called Face_Feature.lib
Before running example, you need to set the Display environment variable and set the display resolution to 640x480.
To perform human face recognition by testing video, you need to do the following:
Test Results:
Test video contains exact same faces for creating features database
Test video is in the middle "focus unclear" state, leading to the lower half of the face, so no face recognition
The face displayed in the lower left of the display is the last recognized face (with matching items)
Only one identifiable face is displayed per frame (even if multiple faces have been identified)
Step 5 - Perform a custom face and USB camera
Further ways to explain whether the example is suitable for custom faces, and video containing the same character but does not have the same facial image.
Test it
Be
Supplement: The item can be modified:
Quality of display surface
Display name except face ID
Display the identified facial image next to each detected facial boundary box
Our other product:
