Simulation 3D Vision Detection Generator

Detect objects and lanes from measurements in 3D simulation environment

expand all in page
  • Library:

  • Automated Driving Toolbox / Simulation 3D

  • Simulation 3D Vision Detection Generator block

Description

The Simulation 3D Vision Detection Generator block generates detections from camera measurements taken by a vision sensor mounted on an ego vehicle in a 3D simulation environment. This environment is rendered using the Unreal Engine® from Epic Games®. The block derives detections from simulated actor poses that are based on cuboid (box-shaped) representations of the actors in the scenario. For more details, see Algorithms.

The block generates detections at intervals equal to the sensor update interval. Detections are referenced to the coordinate system of the sensor. The block can simulate real detections that have added random noise and also generate false positive detections. A statistical model generates the measurement noise, true detections, and false positives. To control the random numbers that the statistical model generates, use the random number generator settings on the Measurements tab of the block.

If you set Sample time to -1, the block uses the sample time specified in the Simulation 3D Scene Configuration block. To use this sensor, you must include a Simulation 3D Scene Configuration block in your model.

Note

The Simulation 3D Scene Configuration block must execute before the Simulation 3D Vision Detection Generator block. That way, the Unreal Engine 3D visualization environment prepares the data before the Simulation 3D Vision Detection Generator block receives it. To check the block execution order, right-click the blocks and select Properties. On the General tab, confirm these Priority settings:

  • Simulation 3D Scene Configuration — 0

  • Simulation 3D Vision Detection Generator — 1

For more information about execution order, see How Unreal Engine Simulation for Automated Driving Works.

Ports

Output

expand all

Object detections, returned as a Simulink bus containing a MATLAB structure. For more details about buses, see Create Nonvirtual Buses (Simulink). The structure has the form shown in this table.

FieldDescriptionType
NumDetectionsNumber of detectionsInteger
IsValidTimeFalse when updates are requested at times that are between block invocation intervalsBoolean
DetectionsObject detectionsArray of object detection structures of length set by the Maximum number of reported detections parameter. Only NumDetections of these detections are actual detections.

The object detection structure contains these properties.

PropertyDefinition
TimeMeasurement time
MeasurementObject measurements
MeasurementNoiseMeasurement noise covariance matrix
SensorIndexUnique ID of the sensor
ObjectClassIDObject classification
ObjectAttributesAdditional information passed to tracker
MeasurementParametersParameters used by initialization functions of nonlinear Kalman tracking filters

The Measurement field reports the position and velocity of a measurement in the coordinate system of the sensor. This field is a real-valued column vector of the form [x; y; z; vx; vy; vz]. The MeasurementNoise field is a 6-by-6 matrix that reports the measurement noise covariance for each coordinate in the Measurement field.

The MeasurementParameters field is a structure that has these fields.

ParameterDefinition
Frame Enumerated type indicating the frame used to report measurements. The Simulation 3D Vision Detection Generator block reports detections in sensor Cartesian coordinates, which is a rectangular coordinate frame. Therefore, for this block, Frame is always set to 'rectangular'.
OriginPositionOffset of the sensor origin from the ego vehicle origin, returned as a vector of the form [x, y, z]. The block derives these values from the x, y, and z mounting position of the sensor. For more details, see the Mounting parameters of this block.
OrientationOrientation of the sensor coordinate frame with respect to the ego vehicle coordinate frame, returned as a 3-by-3 real-valued orthonormal matrix. The block derives these values from the yaw, pitch, and roll mounting orientation of the sensor. For more details, see the Mounting parameters of this block.
HasVelocityIndicates whether measurements contain velocity.

The ObjectClassID property of each detection has a value that corresponds to an object ID. The table shows the object IDs used in the default scenes that you can select from the Simulation 3D Scene Configuration block. If you are using a custom scene, in the Unreal® Editor, you can assign new object types to unused IDs. If a scene contains an object that does not have an assigned ID, that object is assigned an ID of 0. The block detects objects only of class Vehicle, such as vehicles created by using Simulation 3D Vehicle with Ground Following blocks, or of class Road.

IDType
0

None/default

1

Building

2

Not used

3

Other

4

Not used

5

Pole

6

Not used

7

Road

8

Sidewalk

9

Vegetation

10

Vehicle

11

Not used

12

Generic traffic sign

13

Stop sign

14

Yield sign

15

Speed limit sign

16

Weight limit sign

17-18

Not used

19

Left and right arrow warning sign

20

Left chevron warning sign

21

Right chevron warning sign

22

Not used

23

Right one-way sign

24

Not used

25

School bus only sign

26-38

Not used

39

Crosswalk sign

40

Not used

41

Traffic signal

42

Curve right warning sign

43

Curve left warning sign

44

Up right arrow warning sign

45-47

Not used

48

Railroad crossing sign

49

Street sign

50

Roundabout warning sign

51

Fire hydrant

52

Exit sign

53

Bike lane sign

54-56

Not used

57

Sky

58

Curb

59

Flyover ramp

60

Road guard rail

61-66

Not used

67

Deer

68-70

Not used

71

Barricade

72

Motorcycle

73-255

Not used

The ObjectAttributes property of each detection is a structure that has these fields.

FieldDefinition
TargetIndexIdentifier of the actor, ActorID, that generated the detection. For false alarms, this value is negative.

Dependencies

To enable this output port, on the Parameters tab, set the Types of detections generated by sensor parameter to Lanes and objects, Objects only, or Lanes with occlusion.

Lane boundary detections, returned as a Simulink bus containing a MATLAB structure. The structure has these fields.

FieldDescriptionType
TimeLane detection timeReal scalar
IsValidTimeFalse when updates are requested at times that are between block invocation intervalsBoolean
SensorIndexUnique identifier of sensorPositive integer
NumLaneBoundariesNumber of lane boundary detectionsNonnegative integer
LaneBoundariesLane boundary detectionsArray of clothoidLaneBoundary objects

Dependencies

To enable this output port, on the Parameters tab, set the Types of detections generated by sensor parameter to Lanes and objects, Lanes only, or Lanes with occlusion.

Ground truth of actor poses in the simulation environment, returned as a Simulink bus containing a MATLAB structure.

The structure has these fields.

FieldDescriptionType
NumActorsNumber of actorsNonnegative integer
TimeCurrent simulation timeReal-valued scalar
ActorsActor posesNumActors-length array of actor pose structures

Each actor pose structure in Actors has these fields.

FieldDescription
ActorID

Scenario-defined actor identifier, specified as a positive integer.

Position

Position of actor, specified as a real-valued vector of the form [x, y, z]. Units are in meters.

Velocity

Velocity (v) of actor in the x-, y-, and z-direction, specified as a real-valued vector of the form [vx, vy, vz]. Units are in meters per second.

Roll

Roll angle of actor, specified as a real-valued scalar. Units are in degrees.

Pitch

Pitch angle of actor, specified as a real-valued scalar. Units are in degrees.

Yaw

Yaw angle of actor, specified as a real-valued scalar. Units are in degrees.

AngularVelocity

Angular velocity (ω) of actor in the x-, y-, and z-direction, specified as a real-valued vector of the form [ωx, ωy, ωz]. Units are in degrees per second.

The pose of the ego vehicle is excluded from the Actors array.

Dependencies

To enable this output port, on the Ground Truth tab, select the Output actor truth parameter.

Ground truth of lane boundaries in the simulation environment, returned as a Simulink bus containing a MATLAB structure.

The structure has these fields.

FieldDescriptionType
NumLaneBoundariesNumber of lane boundariesNonnegative integer
TimeCurrent simulation timeReal scalar
LaneBoundariesLane boundariesNumLaneBoundaries-length array of lane boundary structures

Each lane boundary structure in LaneBoundaries has these fields.

FieldDescription

Coordinates

Lane boundary coordinates, specified as a real-valued N-by-3 matrix, where N is the number of lane boundary coordinates. Lane boundary coordinates define the position of points on the boundary at specified longitudinal distances away from the ego vehicle, along the center of the road.

  • In MATLAB, specify these distances by using the 'XDistance' name-value pair argument of the laneBoundaries function.

  • In Simulink, specify these distances by using the Distances from ego vehicle for computing boundaries (m) parameter of the Scenario Reader block or the Distance from parent for computing lane boundaries parameter of the Simulation 3D Vision Detection Generator block.

This matrix also includes the boundary coordinates at zero distance from the ego vehicle. These coordinates are to the left and right of the ego-vehicle origin, which is located under the center of the rear axle. Units are in meters.

Curvature

Lane boundary curvature at each row of the Coordinates matrix, specified as a real-valued N-by-1 vector. N is the number of lane boundary coordinates. Units are in radians per meter.

CurvatureDerivative

Derivative of lane boundary curvature at each row of the Coordinates matrix, specified as a real-valued N-by-1 vector. N is the number of lane boundary coordinates. Units are in radians per square meter.

HeadingAngle

Initial lane boundary heading angle, specified as a real scalar. The heading angle of the lane boundary is relative to the ego vehicle heading. Units are in degrees.

LateralOffset

Distance of the lane boundary from the ego vehicle position, specified as a real scalar. An offset to a lane boundary to the left of the ego vehicle is positive. An offset to the right of the ego vehicle is negative. Units are in meters.

BoundaryType

Type of lane boundary marking, specified as one of these values:

  • 'Unmarked' — No physical lane marker exists

  • 'Solid' — Single unbroken line

  • 'Dashed' — Single line of dashed lane markers

  • 'DoubleSolid' — Two unbroken lines

  • 'DoubleDashed' — Two dashed lines

  • 'SolidDashed' — Solid line on the left and a dashed line on the right

  • 'DashedSolid' — Dashed line on the left and a solid line on the right

Strength

Saturation strength of the lane boundary marking, specified as a real scalar from 0 to 1. A value of 0 corresponds to a marking whose color is fully unsaturated. The marking is gray. A value of 1 corresponds to a marking whose color is fully saturated.

Width

Lane boundary width, specified as a positive real scalar. In a double-line lane marker, the same width is used for both lines and for the space between lines. Units are in meters.

Length

Length of dash in dashed lines, specified as a positive real scalar. In a double-line lane marker, the same length is used for both lines.

Space

Length of space between dashes in dashed lines, specified as a positive real scalar. In a dashed double-line lane marker, the same space is used for both lines.

The number of returned lane boundary structures depends on the Maximum number of reported lanes parameter value.

Dependencies

To enable this output port, on the Ground Truth tab, select the Output lane truth parameter.

Parameters

expand all

Mounting

Unique sensor identifier, specified as a positive integer. In a multisensor system, the sensor identifier distinguishes between sensors. When you add a new sensor block to your model, the Sensor identifier of that block is N + 1. N is the highest Sensor identifier value among existing sensor blocks in the model.

Example: 2

Name of the parent to which the sensor is mounted, specified as Scene Origin or as the name of a vehicle in your model. The vehicle names that you can select correspond to the Name parameters of the Simulation 3D Vehicle with Ground Following blocks in your model. If you select Scene Origin, the block places a sensor at the scene origin.

Example: SimulinkVehicle1

Sensor mounting location.

  • When Parent name is Scene Origin, the block mounts the sensor to the origin of the scene. You can set the Mounting location to Origin only. During simulation, the sensor remains stationary.

  • When Parent name is the name of a vehicle (for example, SimulinkVehicle1) the block mounts the sensor to one of the predefined mounting locations described in the table. During simulation, the sensor travels with the vehicle.

Vehicle Mounting LocationDescriptionOrientation Relative to Vehicle Origin [Roll, Pitch, Yaw] (deg)
Origin

Forward-facing sensor mounted to the vehicle origin, which is on the ground, at the geometric center of the vehicle (see Coordinate Systems for Unreal Engine Simulation in Automated Driving Toolbox)

[0, 0, 0]
Front bumper

Forward-facing sensor mounted to the front bumper

[0, 0, 0]
Rear bumper

Backward-facing sensor mounted to the rear bumper

[0, 0, 180]
Right mirror

Downward-facing sensor mounted to the right side-view mirror

[0, –90, 0]
Left mirror

Downward-facing sensor mounted to the left side-view mirror

[0, –90, 0]
Rearview mirror

Forward-facing sensor mounted to the rearview mirror, inside the vehicle

[0, 0, 0]
Hood center

Forward-facing sensor mounted to the center of the hood

[0, 0, 0]
Roof center

Forward-facing sensor mounted to the center of the roof

[0, 0, 0]

Roll, pitch, and yaw are clockwise-positive when looking in the positive direction of the X-axis, Y-axis, and Z-axis, respectively. When looking at a vehicle from the top down, the yaw angle (that is, the orientation angle) is counterclockwise-positive, because you are looking in the negative direction of the axis.

The (X, Y, Z) mounting location of the sensor relative to the vehicle depends on the vehicle type. To specify the vehicle type, use the Type parameter of the Simulation 3D Vehicle with Ground Following block to which you are mounting the sensor. To obtain the (X, Y, Z) mounting locations for a vehicle type, see the reference page for that vehicle.

To determine the location of the sensor in world coordinates, open the sensor block. Then, on the Ground Truth tab, select Output location (m) and orientation (rad) and inspect the data from the Location output port.

Select this parameter to specify an offset from the mounting location by using the Relative translation [X, Y, Z] (m) and Relative rotation [Roll, Pitch, Yaw] (deg) parameters.

Translation offset relative to the mounting location of the sensor, specified as a real-valued 1-by-3 vector of the form [X, Y, Z]. Units are in meters.

If you mount the sensor to a vehicle by setting Parent name to the name of that vehicle, then X, Y, and Z are in the vehicle coordinate system, where:

  • The X-axis points forward from the vehicle.

  • The Y-axis points to the left of the vehicle, as viewed when looking in the forward direction of the vehicle.

  • The Z-axis points up.

The origin is the mounting location specified in the Mounting location parameter. This origin is different from the vehicle origin, which is the geometric center of the vehicle.

If you mount the sensor to the scene origin by setting Parent name to Scene Origin, then X, Y, and Z are in the world coordinates of the scene.

For more details about the vehicle and world coordinate systems, see Coordinate Systems for Unreal Engine Simulation in Automated Driving Toolbox.

Example: [0,0,0.01]

Dependencies

To enable this parameter, select Specify offset.

Rotational offset relative to the mounting location of the sensor, specified as a real-valued 1-by-3 vector of the form [Roll, Pitch, Yaw] . Roll, pitch, and yaw are the angles of rotation about the X-, Y-, and Z-axes, respectively. Units are in degrees.

If you mount the sensor to a vehicle by setting Parent name to the name of that vehicle, then X, Y, and Z are in the vehicle coordinate system, where:

  • The X-axis points forward from the vehicle.

  • The Y-axis points to the left of the vehicle, as viewed when looking in the forward direction of the vehicle.

  • The Z-axis points up.

  • Roll, pitch, and yaw are clockwise-positive when looking in the forward direction of the X-axis, Y-axis, and Z-axis, respectively. If you view a scene from a 2D top-down perspective, then the yaw angle (also called the orientation angle) is counterclockwise-positive because you are viewing the scene in the negative direction of the Z-axis.

The origin is the mounting location specified in the Mounting location parameter. This origin is different from the vehicle origin, which is the geometric center of the vehicle.

If you mount the sensor to the scene origin by setting Parent name to Scene Origin, then X, Y, and Z are in the world coordinates of the scene.

For more details about the vehicle and world coordinate systems, see Coordinate Systems for Unreal Engine Simulation in Automated Driving Toolbox.

Example: [0,0,10]

Dependencies

To enable this parameter, select Specify offset.

Sample time of the block in seconds, specified as a positive scalar. The 3D simulation environment frame rate is the inverse of the sample time.

If you set the sample time to -1, the block inherits its sample time from the Simulation 3D Scene Configuration block.

Parameters

Detection Reporting

Types of detections generated by the sensor, specified as one of these options:

  • Lanes and objects — Detect lanes and objects. No road information is used to occlude actors.

  • Objects only — Detect objects only.

  • Lanes only — Detection lanes only.

  • Lanes with occlusion — Detect lane and objects. Objects in the camera field of view can impair the ability of the sensor to detect lanes.

Maximum number of detections reported by the sensor, specified as a positive integer. Detections are reported in order of increasing distance from the sensor until the maximum number is reached.

Example: 100

Maximum number of reported lanes, specified as a positive integer.

Example: 100

Distances from the parent frame at which to compute the lane boundaries, specified as an N-element real-valued vector. N is the number of distance values. Units are in meters.

The parent is the frame to which the sensor is mounted, such as the ego vehicle. The Parent name parameter determines the parent frame. Distances are relative to the origin of the parent frame.

When detecting lanes from rear-facing cameras, specify negative distances. When detecting lanes from front-facing cameras, specify positive distances.

By default, the block computes a lane boundary every 0.5 meters over the range from 0 to 9.5 meters ahead of the parent.

Example: 1:0.1:10 computes a lane boundary every 0.1 meters over the range from 1 to 10 meters ahead of the parent.

Output Port Settings

Source of object bus name, specified as Auto or Property. If you select Auto, the block creates a bus name. If you select Property, specify the bus name by using the Object bus name parameter.

Object bus name, specified as a valid bus name.

Dependencies

To enable this parameter, set the Source of object bus name parameter to Property.

Source of output lane bus name, specified as Auto or Property. If you select Auto, the block creates a bus name. If you select Property, specify the bus name by using the Specify an output lane bus name parameter.

Lane bus name, specified as a valid bus name.

Dependencies

To enable this parameter, set the Source of output lane bus name parameter to Property.

Measurements

Maximum detection range, specified as a positive real scalar. The vision sensor cannot detect objects beyond this range. Units are in meters.

Example: 250

Object Detector Settings

Bounding box accuracy, specified as a positive real scalar. This quantity defines the accuracy with which the detector can match a bounding box to a target. Units are in pixels.

Example: 9

Noise intensity used for filtering position and velocity measurements, specified as a positive real scalar. Noise intensity defines the standard deviation of the process noise of the internal constant-velocity Kalman filter used in a vision sensor. The filter models the process noise by using a piecewise-constant white noise acceleration model. Noise intensity is typically of the order of the maximum acceleration magnitude expected for a target. Units are in meters per second squared.

Example: 2

Maximum detectable object speed, specified as a nonnegative real scalar. Units are in meters per second.

Example: 20

Maximum allowed occlusion of an object, specified as a real scalar in the range [0 1). Occlusion is the fraction of the total surface area of an object that is not visible to the sensor. A value of 1 indicates that the object is fully occluded. Units are dimensionless.

Example: 0.2

Minimum height and width of an object that the vision sensor detects within an image, specified as a [minHeight,minWidth] vector of positive values. The 2-D projected height of an object must be greater than or equal to minHeight. The projected width of an object must be greater than or equal to minWidth. Units are in pixels.

Example: [25 20]

Probability of detecting a target, specified as a positive real scalar less than or equal to 1. This quantity defines the probability that the sensor detects a detectable object. A detectable object is an object that satisfies the minimum detectable size, maximum range, maximum speed, and maximum allowed occlusion constraints.

Example: 0.95

Number of false detections generated by the vision sensor per image, specified as a nonnegative real scalar.

Example: 1.0

Lane Detector Settings

Minimum size of a projected lane marking in the camera image that the sensor can detect after accounting for curvature, specified as a 1-by-2 real-valued vector of the form [minHeight, minWidth]. Lane markings must exceed both of these values to be detected. Units are in pixels.

Accuracy of lane boundaries, specified as a positive real scalar. This parameter defines the accuracy with which the lane sensor can place a lane boundary. Units are in pixels.

Example: 2.5

Random Number Generator Settings

Select this parameter to add noise to vision sensor measurements. Otherwise, the measurements are noise-free. The MeasurementNoise property of each detection is always computed and is not affected by the value you specify for the Add noise to measurements parameter.

Method to set the random number generator seed, specified as one of the options in the table.

OptionDescription
Repeatable

The block generates a random initial seed for the first simulation and reuses this seed for all subsequent simulations. Select this parameter to generate repeatable results from the statistical sensor model. To change this initial seed, at the MATLAB command prompt, enter: clear all.

Specify seedSpecify your own random initial seed for reproducible results by using the Initial seed parameter.
Not repeatableThe block generates a new random initial seed after each simulation run. Select this parameter to generate nonrepeatable results from the statistical sensor model.

Random number generator seed, specified as a nonnegative integer less than 232.

Example: 2001

Dependencies

To enable this parameter, set the Select method to specify initial seed parameter to Specify seed.

Camera Intrinsics

Camera focal length, in pixels, specified as a two-element real-valued vector. See also the FocalLength (Computer Vision Toolbox) property of cameraIntrinsics.

Example: [480,320]

Optical center of the camera, in pixels, specified as a two-element real-valued vector. See also the PrincipalPoint (Computer Vision Toolbox) property of cameraIntrinsics.

Example: [480,320]

Image size produced by the camera, in pixels, specified as a two-element vector of positive integers. See also the ImageSize (Computer Vision Toolbox) property of cameraIntrinsics.

Example: [240,320]

Radial distortion coefficients, specified as a two-element or three-element real-valued vector. For details on setting these coefficients, see the RadialDistortion (Computer Vision Toolbox) property of cameraIntrinsics.

Example: [1,1]

Tangential distortion coefficients, specified as a two-element real-valued vector. For details on setting these coefficients, see the TangentialDistortion (Computer Vision Toolbox) property of cameraIntrinsics.

Example: [1,1]

Skew angle of the camera axes, specified as a real scalar. See also the Skew (Computer Vision Toolbox) property of cameraIntrinsics.

Example: 0.1

Ground Truth

Select this parameter to output the ground truth of actors on the Actor Truth output port.

Select this parameter to output the ground truth of lane boundaries on the Lane Truth output port.

Model Examples

Simulate Vision and Radar Sensors in Unreal Engine Environment

Simulate Vision and Radar Sensors in Unreal Engine Environment

Implement a synthetic data simulation for tracking and sensor fusion using Simulink and the Unreal Engine simulation environment.

Tips

  • The sensor is unable to detect lanes and objects from vantage points too close to the ground. After mounting the sensor block to a vehicle by using the Parent name parameter, set the Mounting location parameter to one of the predefined mounting locations on the vehicle.

    If you leave Mounting location set to Origin, which mounts the sensor on the ground below the vehicle center, then specify an offset that is at least 0.1 meter above the ground. Select Specify offset, and in the Relative translation [X, Y, Z] (m) parameter, set a Z value of at least 0.1.

  • To visualize detections and sensor coverage areas, use the Bird's-Eye Scope. See Visualize Sensor Data from Unreal Engine Simulation Environment.

  • Because the Unreal Engine can take a long time to start between simulations, consider logging the signals that the sensors output. See Configure a Signal for Logging (Simulink).

Algorithms

To generate detections, the Simulation 3D Vision Detection Generator block feeds the actor and lane ground truth data that is read from the Unreal Engine simulation environment to a Vision Detection Generator block. This block returns detections that are based on cuboid, or box-shaped, representations of the actors. The physical dimensions of detected actors are not based on their dimensions in the Unreal Engine environment. Instead, they are based on the default values set in the Actor Profiles parameter tab of the Vision Detection Generator block, where all actors are approximately the size of a sedan. If you return detections that have occlusions, then the occlusions are based on all actors being of this one size.

See Also

Apps

Blocks

Objects

Topics

Introduced in R2020b

Automated Driving Toolbox Documentation

Support