Skip to content
This repository has been archived by the owner on Jan 7, 2023. It is now read-only.

Latest commit

 

History

History
207 lines (206 loc) · 8.37 KB

video_classification.md

File metadata and controls

207 lines (206 loc) · 8.37 KB
Raw

Classification for Video

This project supports multiple CNN models for classification. Please make sure you have already set up environment and installed this project correctly. You can refer to the following links for your interested models then.

1 CNN Models

2 Run with Other ROS Supported Cameras

3 Other Arguments

1 CNN Models

  • AlexNet

Compile NCS graph.

cd /opt/movidius/ncappzoo/caffe/AlexNet
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=alexnet camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=alexnet camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • GoogleNet

Compile NCS graph.

cd /opt/movidius/ncappzoo/caffe/GoogleNet
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=googlenet camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=googlenet camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • SqueezeNet

Compile NCS graph.

cd /opt/movidius/ncappzoo/caffe/SqueezeNet
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=squeezenet camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=squeezenet camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • Inception_V1

Compile NCS graph.

cd /opt/movidius/ncappzoo/tensorflow/inception_v1
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v1 camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v1 camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • Inception_V2

Compile NCS graph.

cd /opt/movidius/ncappzoo/tensorflow/inception_v2
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v2 camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v2 camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • Inception_V3

Compile NCS graph.

cd /opt/movidius/ncappzoo/tensorflow/inception_v3
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v3 camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v3 camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • Inception_V4

Compile NCS graph.

cd /opt/movidius/ncappzoo/tensorflow/inception_v4
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v4 camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=inception_v4 camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

  • MobileNet

Compile NCS graph.

cd /opt/movidius/ncappzoo/tensorflow/mobilenets
make

Launch video streaming nodelet.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=mobilenet camera:=realsense
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=mobilenet camera:=usb

Launch image viewer to show the classification result on another console.

# Option 1: RealSense
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/camera/color/image_raw"
# Option 2: USB camera
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:="/usb_cam/image_raw"

2 Run with Other ROS Supported Cameras

Though this ROS package is only tested on RealSense D400 series camera and Microsoft HD-300 USB camera, other ROS supported cameras probably work as well. You can try in this way:
Launch your preferred camera node.

#launch ROS master
roscore
#launch camera node on another console
rosrun <your-camera-pkg> <your-camera-node>

Launch video streaming nodelet and assign input_topic with the topic URL of your RGB camera.

# Launch video streaming nodelet
roslaunch movidius_ncs_launch ncs_camera.launch cnn_type:=googlenet camera:=others input_topic:=<your_rgb_camera_topic>
# Launch image viewer to show the classification result on another console
roslaunch movidius_ncs_launch ncs_stream_classification_example.launch camera_topic:=<your_rgb_camera_topic>

3 Other Arguments

Arguments Description Default Value Valid Values
device_index ncs device index 0 0~N-1(N is the maximum number of inserted NCS devices)
log_level ncs log level 0 0:Nothing / 1:Errors / 2:Verbose
cnn_type indicate different cnn models googlenet alexnet / googlenet / squeezenet / inception_v1 / inception_v2 / inception_v3 / inception_v4 / mobilenet
top_n the number of results to be shown, only valid for classification 3 0~5
input_topic RGB camera topic to be subscribed /camera/color/image_raw only valid when "camera:=others"
output_topic published topic of inference results /movidius_ncs_nodelet/classified_objects any valid topic name
camera camera type others usb / realsense / others
color_width frame width of RealSense D400 640 any resolution that RealSense D400 supports
color_height frame height of RealSense D400 480 any resolution that RealSense D400 supports
image_width frame width of USB camera 640 any resolution that USB camera supports
image_height frame height of USB camera 480 any resolution that USB camera supports
video_device USB camera device node /dev/video0 any available video device node