CloudSoftwareHardware
Software Stack
DepthAI
OAK Apps
AI Models
Troubleshooting

ON THIS PAGE

  • FeatureTracker
  • How to place it
  • Inputs and Outputs
  • Usage
  • Examples of functionality
  • How it works
  • Image cells
  • Initial Harris Threshold
  • Entry conditions for new features
  • Harris Threshold for Tracked Features
  • Feature Maintenance
  • New position calculation
  • Reference

FeatureTracker

Supported on:RVC2RVC4
FeatureTracker node detects key points (features) on a frame and tracks them on the next frame. The valid features are obtained from the Harris score or Shi-Tomasi. The default number of target features is 320 and the default maximum number of features is 480.

How to place it

Python
C++

Python

Python
1with dai.Pipeline() as pipeline:
2    featureTracker = pipeline.create(dai.node.FeatureTracker)

Inputs and Outputs

React Flow

Usage

Python
C++

Python

Python
1with dai.Pipeline() as pipeline:
2    featureTracker = pipeline.create(dai.node.FeatureTracker)
3
4    # Set number of shaves and number of memory slices to maximum
5    featureTracker.setHardwareResources(2, 2)
6    # Specify to wait until configuration message arrives to inputConfig Input.
7    featureTracker.setWaitForConfigInput(True)
8
9    # You have to use Feature tracker in combination with
10    # an image frame source - mono/color camera or xlinkIn node

Examples of functionality

How it works

Image cells

To have features all around the image, it is divided into cells which are then processed separately. Each cell has a target feature count = frame target features / number of cells. The number of cells can be configured in horizontal and in vertical direction. The default number of cells is 4 (horizontal) x 4 (vertical). This means that the default number of target features per cell is: 320 / (4 * 4) = 20. Note that if an already tracked point happens to have its new coordinate in a full cell, it will not be removed, therefore number of features can exceed this limit.

Initial Harris Threshold

This threshold controls the minimum strength of a feature which will be detected. Setting this value to 0 enables the automatic thresholds, which are used to adapt to different scenes. If there is a lot of texture, this value needs to be increased, to limit the number of detected points. Each cell has its own threshold.

Entry conditions for new features

The entry conditions for new features are:
  • features must not be too close to each other (minimum distance criteria - default value is 50, the unit of measurement being squared euclidean distance in pixels),
  • Harris score of the feature is high enough,
  • there is enough room in the cell for the feature (target feature count is not achieved).

Harris Threshold for Tracked Features

Once a feature was detected and we started tracking it, we need to update its Harris score on each image. This threshold defines the point where such a feature must be removed. The goal is to track points for as long as possible, so the conditions for entry points are higher than the ones for already tracked points. This is why, this value is usually smaller than the detection threshold.

Feature Maintenance

The algorithm has to decide which feature will be removed and which will be kept in the subsequent frames. Note that tracked features have priority over new features. It will remove the features which:
  • have too large tracking error (wasn't tracked correctly),
  • have too small Harris score (configurable threshold).

New position calculation

A position of the previous features on the current frame can be calculated in two ways:
  1. Using the pyramidal Lucas-Kanade optical flow method.
  2. Using a dense motion estimation hardware block (Block matcher).

Reference

class

dai::node::FeatureTracker

#include FeatureTracker.hpp
variable
std::shared_ptr< FeatureTrackerConfig > initialConfig
Initial config to use for feature tracking.
variable
Input inputConfig
Input FeatureTrackerConfig message with ability to modify parameters in runtime. Default queue is non-blocking with size 4.
variable
Input inputImage
Input message with frame data on which feature tracking is performed. Default queue is non-blocking with size 4.
variable
Output outputFeatures
Outputs TrackedFeatures message that carries tracked features results.
variable
Output passthroughInputImage
Passthrough message on which the calculation was performed. Suitable for when input queue is set to non-blocking behavior.
function
FeatureTracker()
function
FeatureTracker(std::unique_ptr< Properties > props)
function
void setHardwareResources(int numShaves, int numMemorySlices)
Specify allocated hardware resources for feature tracking. 2 shaves/memory slices are required for optical flow, 1 for corner detection only.
Parameters
  • numShaves: Number of shaves. Maximum 2.
  • numMemorySlices: Number of memory slices. Maximum 2.
inline function
DeviceNodeCRTP()
inline function
DeviceNodeCRTP(const std::shared_ptr< Device > & device)
inline function
DeviceNodeCRTP(std::unique_ptr< Properties > props)
inline function
DeviceNodeCRTP(std::unique_ptr< Properties > props, bool confMode)
inline function
DeviceNodeCRTP(const std::shared_ptr< Device > & device, std::unique_ptr< Properties > props, bool confMode)

Need assistance?

Head over to Discussion Forum for technical support or any other questions you might have.