Object Detection#

About#

This feature lets you generate object detection using existing cameras in AirSim, similar to detection DNN.
Using the API you can control which object to detect by name and radius from camera.
One can control these settings for each camera, image type and vehicle combination separately.

API#

  • Set mesh name to detect in wildcard format
    simAddDetectionFilterMeshName(camera_name, image_type, mesh_name, vehicle_name = '')

  • Clear all mesh names previously added
    simClearDetectionMeshNames(camera_name, image_type, vehicle_name = '')

  • Set detection radius in cm
    simSetDetectionFilterRadius(camera_name, image_type, radius_cm, vehicle_name = '')

  • Get detections
    simGetDetections(camera_name, image_type, image_requests, vehicle_name = '')

The return value of simGetDetections is a tuple of DetectionInfo array and ImageResponse. Below is an explanation of the properties of DetectionInfo:

DetectionInfo
    name = ''  # Specifies the instance name associated with the detection.
    geo_point = GeoPoint()  # Represents the geographical point associated with the detection.
    box2D = Box2D()  # Represents the 2D bounding box enclosing the detected object on the projected image.
    box3D = Box3D()  # Represents the 3D bounding box enclosing the detected object, referenced by the camera of the drone.
    relative_pose = Pose()  # Represents the relative pose of the detected object referenced by the camera of the drone.

Usage example#

Python script detection.py shows how to set detection parameters and shows the result in OpenCV capture.

A minimal example using API with Blocks environment to detect Cylinder objects: 

camera_name = "0"
image_type = airsim.ImageType.Scene

client = airsim.MultirotorClient()
client.confirmConnection()

client.simSetDetectionFilterRadius(camera_name, image_type, 80 * 100) # in [cm]
client.simAddDetectionFilterMeshName(camera_name, image_type, "Cylinder_*") 
client.simGetDetections(camera_name, image_type)
detections = client.simClearDetectionMeshNames(camera_name, image_type)

Output result: 

Cylinder: <DetectionInfo> {   'box2D': <Box2D> {   'max': <Vector2r> {   'x_val': 617.025634765625,
    'y_val': 583.5487060546875},
    'min': <Vector2r> {   'x_val': 485.74359130859375,
    'y_val': 438.33465576171875}},
    'box3D': <Box3D> {   'max': <Vector3r> {   'x_val': 4.900000095367432,
    'y_val': 0.7999999523162842,
    'z_val': 0.5199999809265137},
    'min': <Vector3r> {   'x_val': 3.8999998569488525,
    'y_val': -0.19999998807907104,
    'z_val': 1.5199999809265137}},
    'geo_point': <GeoPoint> {   'altitude': 16.979999542236328,
    'latitude': 32.28772183970703,
    'longitude': 34.864785008379876},
    'name': 'Cylinder9_2',
    'relative_pose': <Pose> {   'orientation': <Quaternionr> {   'w_val': 0.9929741621017456,
    'x_val': 0.0038591264747083187,
    'y_val': -0.11333247274160385,
    'z_val': 0.03381215035915375},
    'position': <Vector3r> {   'x_val': 4.400000095367432,
    'y_val': 0.29999998211860657,
    'z_val': 1.0199999809265137}}}

image image