background_subtraction

Separates the foreground from the background with the background subtractors of OpenCV: MOG2 and KNN, and MOG, GMG, CNT, GSOC and LSBP of opencv_contrib.

opencv_apps	3.0.0 and later
ROS	ROS One, jazzy, kilted, lyrical and later
OpenCV	the bgsegm module of opencv_contrib for MOG, GMG, CNT, GSOC and LSBP
Group	Segmentation, the nodes of which publish the mask or label image, `~/image` of the color filters and background_subtraction, `~/label` of dnn_semantic_segmentation, and `~/contours` (ContourArrayStamped) of the regions of segment_objects and watershed_segmentation

Usage

# ROS 2
ros2 launch opencv_apps background_subtraction.launch.xml image:=/camera/image_raw debug_view:=false
# ROS 1
roslaunch opencv_apps background_subtraction.launch image:=/camera/image_raw debug_view:=false

Try it with the demo video

Download the input video of the demo above, publish it as image with image_publisher, and run the node on it, in the terminals:

wget https://ros-perception.github.io/opencv_apps/media/clips/pool.mp4
# terminal 1: publish the video as image in a loop
ros2 run image_publisher image_publisher_node pool.mp4 --ros-args -p publish_rate:=10.0 -r image_raw:=image
# terminal 2: run the node with the parameters of the demo
ros2 launch opencv_apps background_subtraction.launch.xml image:=image method:=0 node_name:=background_subtraction_mog2
# terminal 3: view the result
ros2 run rqt_image_view rqt_image_view /background_subtraction_mog2/image

MOG2; the camera is fixed and the balls move.

In ROS 1:

# terminal 1
rosrun image_publisher image_publisher pool.mp4 _publish_rate:=10 __name:=camera
# terminal 2
roslaunch opencv_apps background_subtraction.launch image:=camera/image_raw method:=0 node_name:=background_subtraction_mog2

See Credits of the demo videos for the license of the video.

Subscribed topics

Topic	Type
`image`	`sensor_msgs/msg/Image`

Published topics

Topic	Type
`~/background`	`sensor_msgs/msg/Image`
`~/image`	`sensor_msgs/msg/Image`

Parameters

See also the common parameters.

Name	Type	Default (range)	Changed at run time	Description
`cnt_max_pixel_stability`	int	900 (1 - 10000)	yes	CNT: Maximum allowed credit for a pixel in history
`cnt_min_pixel_stability`	int	15 (1 - 1000)	yes	CNT: Number of frames with same pixel color to consider stable
`cnt_use_history`	bool	true	yes	CNT: Give a pixel credit for being stable for a long time
`gmg_background_prior`	double	0.8 (0 - 1)	yes	GMG: Prior probability that any given pixel is a background pixel
`gmg_decision_threshold`	double	0.8 (0 - 1)	yes	GMG: Threshold of the probability above which a pixel is foreground
`gmg_initialization_frames`	int	120 (1 - 1000)	yes	GMG: Number of frames used to initialize the background models
`gmg_max_features`	int	64 (1 - 512)	yes	GMG: Total number of distinct colors to maintain in histogram (recreates the model)
`gmg_quantization_levels`	int	16 (1 - 128)	yes	GMG: Number of discrete levels in each channel to be used in histograms (recreates the model)
`gmg_smoothing_radius`	int	7 (0 - 31)	yes	GMG: Kernel size of the median filter to clean up the foreground mask, 0 (no filter) or odd
`gsoc_hits_threshold`	int	32 (1 - 1000)	yes	GSOC: How many positives the sample must get before it will be considered as a possible replacement (recreates the model)
`gsoc_n_samples`	int	20 (2 - 100)	yes	GSOC: Number of samples to maintain at each point of the frame (recreates the model)
`gsoc_propagation_rate`	double	0.01 (0 - 1)	yes	GSOC: Probability of propagating to neighbors (recreates the model)
`gsoc_replace_rate`	double	0.003 (0 - 1)	yes	GSOC: Probability of replacing the old sample (recreates the model)
`knn_detect_shadows`	bool	true	yes	KNN: Detect shadows and mark them (127) in the foreground mask
`knn_dist2_threshold`	double	400.0 (1 - 5000)	yes	KNN: Threshold on the squared distance between the pixel and the sample to decide whether a pixel is close to that sample
`knn_history`	int	500 (1 - 5000)	yes	KNN: Length of the history
`learning_rate`	double	-1.0 (-1 - 1)	yes	Learning rate of the background model, negative: automatically chosen, 0: not updated, 1: reinitialized from the last frame (CNT, GSOC and LSBP may still update the model at 0, and GMG, GSOC and LSBP do not reinitialize it at 1)
`lsbp_min_count`	int	2 (1 - 100)	yes	LSBP: Minimal number of matches for a pixel to be background (recreates the model)
`lsbp_n_samples`	int	20 (2 - 100)	yes	LSBP: Number of samples to maintain at each point of the frame (recreates the model)
`lsbp_radius`	int	16 (1 - 64)	yes	LSBP: Radius of the LSBP descriptor (recreates the model)
`lsbp_threshold`	int	8 (0 - 32)	yes	LSBP: Threshold of the LSBP descriptor distance (recreates the model)
`method`	int	0 (0 - 6)	yes	Background subtraction method, 0: MOG2, 1: KNN, 2: MOG, 3: GMG, 4: CNT, 5: GSOC, 6: LSBP (MOG, GMG, CNT, GSOC and LSBP need the bgsegm module of opencv_contrib, otherwise MOG2 is used)
`mog2_detect_shadows`	bool	true	yes	MOG2: Detect shadows and mark them (127) in the foreground mask
`mog2_history`	int	500 (1 - 5000)	yes	MOG2: Length of the history
`mog2_nmixtures`	int	5 (1 - 10)	yes	MOG2: Number of Gaussian components in the background model (recreates the model)
`mog2_var_threshold`	double	16.0 (1 - 256)	yes	MOG2: Threshold on the squared Mahalanobis distance between the pixel and the model to decide whether a pixel is well described by the background model
`mog_background_ratio`	double	0.7 (0 - 1)	yes	MOG: Background ratio
`mog_history`	int	200 (1 - 5000)	yes	MOG: Length of the history
`mog_nmixtures`	int	5 (1 - 8)	yes	MOG: Number of Gaussian mixtures, at most 8 by OpenCV (recreates the model)
`mog_noise_sigma`	double	0.0 (0 - 100)	yes	MOG: Noise strength, 0: automatic
`queue_size`	int	3	no, restart the node	Specify queue_size of input image subscribers

Source

src/nodelet/background_subtraction_nodelet.cpp, launch/background_subtraction.launch.xml