andreasjakl.com

In the previous parts, we’ve taken a look behind the scenes and manually implemented a depth map with Python and OpenCV. Now, let’s compare the results to Unity’s AR Foundation.

How exactly do depth maps work in ARCore? While Google’s paper describes their approach in detail, their implementation is not open source.

However, Google has released a sample project along with a further paper called DepthLab . It’s directly accessing the ARCore depth API and builds complete sample use-cases on top of them.

DepthLab is available as an open-source Unity app. They use the ARCore SDK for Unity directly and not yet the AR Foundation package.

Depth Maps with AR Foundation in Unity

However, Google recommends using AR Foundation with their own ARCore Extensions module (if needed; currently, they only add Cloud Anchor support). Therefore, let’s take a closer look at how to create depth maps using ARFoundation.

In part 2, we rectified our two camera images. The last major step is stereo matching. The algorithm that Google is using for ARCore is an optimized hybrid of two previous publications: PatchMatch Stereo and HashMatch .

An implementation in OpenCV is based on Semi-Global Matching (SGM) as published by Hirschmüller . In Google’s paper , they compare themselves to an implementation of Hirschmüller and outperform those; but for the first experiments, OpenCV’s default is good enough and provides plenty of room for experimentation.

3. Stereo Matching for the Disparity Map (Depth Map)

OpenCV documentation includes two examples that include the stereo matching / disparity map generation: stereo image matching and depth map.

Most of the following code in this article is just an explanation of the configuration options based on the documentation. Setting fitting values for the scenes you expect is crucial to the success of this algorithm. Some insights are listed in the Choosing Good Stereo Parameters article. These are the most important settings to consider:

• Block size: if set to 1, the algorithm matches on the pixel level. Especially for higher resolution images, bigger block sizes often lead to a cleaner result.
• Minimum / maximum disparity: this should match the expected movements of objects within the images. In freely moving camera settings, a negative disparity could occur as well – when the camera doesn’t only move but also rotate, some parts of the image might move from left to right between keyframes, while other parts move from right to left.
• Speckle: the algorithm already includes some smoothing by avoiding small speckles of different depths than their surroundings.

Visualizing Results of Stereo Matching

I’ve chosen values that work well for the sample images I have captured. After configuring these values, computing the disparity map is a simple function call supplying both rectified images.

In part 1 of the article series, we’ve identified the key steps to create a depth map. We have captured a scene from two distinct positions and loaded them with Python and OpenCV. However, the images don’t line up perfectly fine. A process called stereo rectification is crucial to easily compare pixels in both images to triangulate the scene’s depth!

For triangulation, we need to match each pixel from one image with the same pixel in another image. When the camera rotates or moves forward / backward, the pixels don’t just move left or right; they could also be found further up or down in the image. That makes matching difficult.

Wrapping Images for Stereo Rectification

Image rectification wraps both images. The result is that they appear as if they have been taken only with a horizontal displacement. This simplifies calculating the disparities of each pixel!

With smartphone-based AR like in ARCore, the user can freely move the camera in the real world. The depth map algorithm only has the freedom to choose two distinct keyframes from the live camera stream. As such, the stereo rectification needs to be very intelligent in matching & wrapping the images!

In more technical terms, this means that after stereo rectification, all epipolar lines are parallel to the horizontal axis of the image.

To perform stereo rectification, we need to perform two important tasks:

1. Detect keypoints in each image.
2. We then need the best keypoints where we are sure they are matched in both images to calculate reprojection matrices.
3. Using these, we can rectify the images to a common image plane. Matching keypoints are on the same horizontal epipolar line in both images. This enables efficient pixel / block comparison to calculate the disparity map (= how much offset the same block has between both images) for all regions of the image (not just the keypoints!).

Google’s research improves upon the research performed by Pollefeys et al. . Google additionally addresses issues that might happen, especially in mobile scenarios.

For a realistic Augmented Reality (AR) scene, a depth map of the environment is crucial: if a real, physical object doesn’t occlude a virtual object, it immediately breaks the immersion.

Of course, some devices already include specialized active hardware to create real-time environmental depth maps – e.g., the Microsoft HoloLens or the current high-end iPhones with a Lidar sensor. However, Google decided to go into a different direction: its aim is to bring depth estimation to the mass market, enabling it even for cheaper smartphones that only have a single RGB camera.

In this article series, we’ll look at how it works by analyzing the related scientific papers published by Google. I’ll also show a Python demo based on commonly used comparable algorithms which are present in OpenCV. In the last step, we’ll create a sample Unity project to see depth maps in action. The full Unity example is available on GitHub.

Quick Overview: ARCore Depth Map API

How do Depth Maps with ARCore work? The smartphone saves previous images from the live camera feed and estimates the phone’s motion between these captures. Then, it selects two images that show the same scene from a different position. Based on the parallax effect (objects nearer to you move faster than these farther away – e.g., trees close to a train track move fast versus the mountain in the background moving only very slowly), the algorithm then calculates the distance of this area in the image.

This has the advantage that a single-color camera is enough to estimate the depth. However, this approach needs structured surfaces to detect the movement of unique features in the image. For example, you couldn’t get many insights from two images of a plain white wall, shot from two positions 20 cm apart. Additionally, it’s problematic if the scene isn’t static and objects move around.

As such, given that you have a well-structured and static scene, the algorithm developed by Google works best in a range between 0.5 and 5 meters.