Overview

Bottlenose cameras comes with two feature point detectors: FAST (Features from accelerated segment test) and GFTT (Good Features To Track). Feature points are also referred to as keypoints or corners.

FAST

The FAST algorithm was introduced as a high-speed corner detector. For a given input intensity image, Bottlenose uses the following algorithm to compute feature points:

1. Select a pixel p in the image.
2. Consider a circle of 16 pixels centered around pixel p.
3. If at least 9 contiguous pixels in the circle of 16 pixels have an intensity difference with p greater than a given threshold value, pixel p is declared as a corner.
4. If needed, apply a non-maximum suppression algorithm to filter multiple adjacent corners.

Refer to the original paper, by Rosten and Drummond ECCV'06, for more details on the FAST algorithm.

GFTT

Besides the FAST detector, Bottlenose can also generate features based on the well-known good features to track work by J. Shi and C. Tomasi, CVPR'94. For a given input intensity image, Bottlenose uses the following algorithm to detect good features to track:

1. calculates the corner quality measure for each pixel using either the minimum eigenvalue or the Harris detector method, if so specified.
2. determines a threshold value computed from a set quality level and the maximum of the minimum eigenvalue image.
3. applies a non-maximum suppression on the detected corners.
4. filters the corners to reject corners with an intensity value greater than the computed threshold.
5. orders the remaining corners in descending order
6. throws away each corner for which there is a stronger corner at a distance less than a defined minimum distance

Enabling Feature Point Detection

Bottlenose detects and transmits feature points through the chunk data interface of its GigE Vision protocol. The following steps show how to enable feature points with either the Bottlenose Stereo Viewer or eBusPlayer.

1. Launch Stereo Viewer or eBusPlayer and connect to the camera
2. Click on Device Control to open the control panel
3. Navigate to ChunkDataControl
4. Set ChunkActiveMode to True to turn on chunk data transfer
5. Set ChunkSelector to FeaturePoints to activate feature points
6. Finally, set ChunkEnable to True to enable the feature point detection and transmission

Feature Point Parameters

Bottlenose exposes various parameters that can be used to influence the way it detects feature points using either FAST or GFTT. To access these parameters, navigate to KeypointParameters under the FeatureMacthing section. You may need to set Visibility to Guru to access these settings.

Here is a brief description of each parameter:

• KPCornerType: determines the feature points detector to use, FAST or GFTT.
• KPDetector: available only for GFTT. Choose between Harris and minimum eigenvalue as an underlying detector for GFTT.
• KPMaxNumber: Maximum number of feature points to output. This can be up to 65535 for FAST and 8192 for GFTT.
• KPQualityLevel: Quality level used by GFTT to threshold feature points. Bottlenose computes a threshold t using the maximum of the minimum eigenvalue using the equation t = max_value x (quality_level/1023). Thus, the number of points detected decreases as the quality level increases.
• KPMinimumDistance: Determines the minimum Euclidean distance between two feature points. Available only for GFTT.
• KPHarrisParamK: is the free parameter of the Harris corner detector. This parameter is ignored if the GFTT detector is not Harris.
• KPThreshold: Pixel value threshold for the FAST algorithm.
• KPUseNMS: if set to True, FAST applies non-maximum suppression to filter redundant feature points.

Receiving Feature Points with Python

Feature points can be streamed from Bottlenose using Python. For details on how to interact with the camera and stream feature point data, refer to our sample code.

The following code snippet activates the chunk mode and enables the transmission of feature points.

# Get device parameters from the Gige Vision device
device_params = device.GetParameters()

# Activate chunk mode
chunkMode = device_params.Get("ChunkModeActive")
chunkMode.SetValue(True)

# Select Feature points 
chunkSelector = device_params.Get("ChunkSelector")
chunkSelector.SetValue("FeaturePoints")

# Enable feature points chunk
chunkEnable = device_params.Get("ChunkEnable")
chunkEnable.SetValue(True)

Receiving Feature Points with ROS2

The ROS2 driver for Bottlenose has support for feature points. This feature can be enabled when you launch the node as in the following.

ros2 run bottlenose_camera_driver bottlenose_camera_driver_node --ros-args \
    -p mac_address:="<MAC>" \ 
    -p stereo:=true \
    -p feature_points:=fast9 \
    -p features_threshold:=10    

This command will turn on the stereo acquisition and use FAST as the feature detector with a threshold of 10. All other image processing parameters will be set to their defaults. This node will then publish 4 topics:

1. image_color and image_color_1 for the left and right color images, respectively
2. features and features_1 an ImageMarker2D array of feature points from the left and right images.