Hacky Hour Highlights 33: Exciting New Features

AprilTags, Kalman Tracking, middleware updates

Future Hacky Hours

Recently, we’ve changed our Hacky Hour schedule from weekly to a periodic schedule when we have new feature releases and product updates to edgeIQ. Hacky Hours are a great way to learn introductory content and participate in code demos and engaging exercises led by computer vision experts. We are here to ensure the success of our developer community. In these sessions, our community can ask us questions about setting up an edge device, building a CV application, or general inquiries related to computer vision and alwaysAI!

What's New with edgeIQ 1.3.0

In this Hacky Hour, Steve Griset, the CTO of alwaysAI, and the edgeIQ team, introduced AprilTags, Kalman Tracking, and our recent enhancements to our middleware. Follow along as Steve Griset, Steve Bottos, Machine Learning Engineer, and Taiga Ishida, Software Engineer, cover the specific functions of our latest releases and how they are used in real-world applications. These new updates to our middleware enable our community to work with up-to-date software and train and deploy computer vision applications with new and exciting capabilities. edgeIQ 1.3.0 is included in the installation package of the alwaysAI desktop app and consists of the following new features and enhancements:

  • Upgrade OpenCV to 4.5.1 and OpenVINO to 2021.2​
  • Upgrade to Python 3.7​
  • Speedup YOLOv3 post-processing on DNN backend​
  • Improve OAK camera API
  • AprilTags
  • Kalman Tracker

Kalman Tracker

Steve Bottos, Machine Learning Engineer at alwaysAI, introduces the Kalman filter and its specific functions. Kalman filter is a classic signal processing algorithm commonly applied in tracking scenarios. These filters were designed to model an object's noise-free motion characteristics (position, velocity, acceleration), and predict its next movements. It can also use predictions to account for missing bounding boxes. Common applications include navigation and control of vehicles. 

To compare, let's first consider the Centroid tracker. The Centroid tracker uses information carried over from the last frame in order to match and track objects from one frame to another. It considers only which object from the last frame is closest to an object in the present frame in order to associate objects from frame to frame. The Kalman filter is slightly more heavy-weight than the Centroid tracker, but more suitable for difficult tracking situations. Kalman filters consider predicted future points based on the previous frame's measurements and motion inertia in its decisions.

Compare Centroid Tracker and Kalman Tracker

AprilTags

Taiga Ishida, Software Engineer at alwaysAI, introduces AprilTags as a type of fiducial marker, consisting of a pattern generated on a black square with a white foreground and a black border. The black border makes it easier for image processing algorithms to detect the AprilTags in a wide variety of scenarios, including variations in rotation, scale, and lighting conditions. 

AprilTags are commonly used for:

  • Camera calibration
  • Object size estimation
  • Measuring the distance between the camera and an object
  • 3D positioning
  • Object orientation
  • Robotics (autonomous navigation)

You can conceptually think of the function of AprilTags as similar to QR codes — a 2D binary pattern that can be detected using computer vision algorithms. The significant difference being that AprilTags store less data than QR codes, enabling them to be more easily detected at longer ranges than QR codes.

List of  AprilTags

 

Guest Questions


QUESTION: Do you think a Kalman Tracker is effective for tracking people that are moving chaotically in a crowd?

ANSWER (Steve Bottos): Yes, there's a tracker for each object, and the tracker will be able to eventually converge and understand what each object is capable of and how they behave. It really depends on your detection and the FPS. If you want accurate tracking, you need an accurate object detector, such as an Xavier, or else tracking is just going to fall apart no matter what you use.


QUESTION: Are AprilTags similar to ArUko Markers? And If not what are the differences between them?

ANSWER (Steve): ArUko markers and AprilTags are very similar, for more information visit Robotics Beta.

Check out the GitHub Repo for this Hacky Hour here.

See below for the full video of this Hacky Hour, or click here.

Join us for our upcoming Hacky Hours! Whether you are new to the community or an experienced user of alwaysAI, you are welcome to join, ask questions, and provide the community with information about what you're working on. Register here.

 

Get started now

We are providing professional developers and enterprises with a simple and easy-to-use platform to build and deploy computer vision applications on edge devices. 

Sign Up for Free