Welcome to USB Industrial Grade AI Vision Camera repository

6.Face recognition

Face recognition

1.Overview

wiki：http://wiki.ros.org/opencv_apps

Source code：https://github.com/ros-perception/opencv_apps.git

Most of the code was originally taken from:https://github.com/Itseez/opencv/tree/master/samples/cpp

Functional package：~/software/library_ws/src/opencv_apps

The topic subscribed by this function package is [/image]. What we have to do is to open the camera node, write a topic that converts the camera topic into a [/image] node, and publish the [/image] topic.

Turn on the camera and publish the path of the node of the [/image] topic:


~/yahboomcar_ws/src/yahboomcar_visual/scripts/pub_image.py

The opencv_apps program provides various nodes that internally run the functions of opencv and publish the results to a ROS topic. When using the opencv_apps program, you only need to run a launch file according to your own business needs, so that you no longer have to write program codes for these functions.

ROS Wiki has related node analysis, topic subscription and topic publishing of the corresponding node, introduction of related parameters, etc. See the ROS WiKi for details.

2. Use

Code path: ~/yahboomcar_ws/src/opencv_apps/launch

Start the camera


xxxxxxxxxx
roslaunch yahboomcar_visual opencv_apps.launch img_flip:=false

img_flip parameter: whether the image needs to be flipped horizontally, the default is false.

The [usb_cam-test.launch] file opens the [web_video_server] node by default, and you can directly use the [IP:8080] web page to view images in real time.

Start up Opencv.apps function


xxxxxxxxxx
roslaunch opencv_apps face_recognition.launch             # face recognition

The debug_view for some functions is turned off. If no screen appears, you can view the effect in the following two ways.


xxxxxxxxxx
The reason for closing debug_view is that it will generate errors on the terminal, but the actual effect has not been affected.

Local view screen

Enter the following command and select the corresponding topic to see the effect.


xxxxxxxxxx
rqt_image_view

LAN viewing screen

In the same local area network, enter IP+port (8080) in the browser, for example:


xxxxxxxxxx
192.168.2.150:8080 #IP refers to the host IP address

3. Face recognition display

This case is based on autonomous training and real-time recognition by collecting images of people in real time, and the steps are slightly complicated.

Parameter	Type	Default	Analyze
~approximate_sync	bool	false	Subscribe to the topic [camera_info] to get the default coordinate system ID, otherwise use the image information directly.
~queue_size	int	100	Queue size for subscribing topics
~model_method	string	"eigen"	Face recognition method: "eigen", "fisher" or "LBPH"
~use_saved_data	bool	true	Load training data from ~data_dir path
~save_train_data	bool	true	Save training data to ~data_dir path for retraining
~data_dir	string	"~/opencv_apps/face_data"	Save training data path
~face_model_width	int	190	Width of training face images
~face_model_height	int	90	Training face image height
~face_padding	double	0.1	Filling ratio of each face
~model_num_components	int	0	The number of components of the face recognizer model (0 is considered unlimited)
~model_threshold	double	8000.0	Face recognition model threshold
~lbph_radius	int	1	Radius parameter (for LBPH method only)
~lbph_neighbors	int	8	Neighborhood parameters (only for LBPH method）
~lbph_grid_x	int	8	Grid x parameter (only for LBPH method)
~lbph_grid_y	int	8	Grid y parameter (only for LBPH method)
~queue_size	int	100	Image subscriber queue size

Operation steps:

First, enter the character’s name after the colon in the picture below: Yahboom
Confirm name: y
Then place the face in the center of the image and click OK.
Add a photo in a loop: y, click to confirm.
To end image collection, enter: n and click Confirm.
Close the launch file and restart.

If you need to enter the recognized identifications, cycle through steps 1 to 5 until all identified persons have been entered, and then proceed to step six.

Step 3: Ensure that faces can be recognized

Final recognition effect

ROS+Opencv basics

harris corner detection