This site has different types of tech blogs, including AR, CV, SLAM, Math, Coding and software developments. Each blog will be updated from time to time.
Index of main posts are as following.
More posts could be found in categories tab.
Majority content of this post came from the book Basic Knowledge on Visual SLAM: From Theory to Practice.
SLAM (Simultaneous Localization and Mapping) is the technique to establish the nearby environment and localize the moving object inside it at the same time according to sensor measurements. SLAM can be used in augmented reality, autonomous vehicles, and robot’s navigation. The mathematic theory of SLAM is the state estimation.
Visual SLAM is a kind of SLAM that depends on on visual sensors (cameras), which also involves photometric and geometric computer vision.
Literature review for SLAM papers that use neural network / deep learning based approaches.
Literature review for neural network based reconstruction approach.
The SLAM algorithm based on traditional computer vision geometry and state estimation has been mature. There are lots of open source SLAM architecture, and different commercial products have their own implementation. Every SLAM implementation has its pros and cons. This post is to provide the high level guidance for how to design a practical SLAM system for products.
3D Reconstruction is the technique to recover the scene geometry with camera or other sensor’s measurements.
The classic approaches constructs the meshes (triangles) to describe the surface of the environment, while the advanced neural network could describe the scene with implicit representations.
Literature review for SLAM papers.
Notes of the book Introduction to Linear Algebra.
Object detection is kind of classificitaion which you need to figure out where the object is located in the image. The output is usually described as:
$$
y = [P_c, P_x, P_y, W, H, c_1, c_2, c_3, …]
$$
$P_c$ means whether this image contain the content you’re detecting, $c_i$ the probability of the label_i object exists.
Object detection includes both localization and detection.
$$
Intersection\ Over\ Union = \frac{size\ of\ overlaping\ area}{size\ of\ true\ and\ detection\ area}
$$
Introduced classic deep learning model structures.