Learning image recognition processing, want to get a higher ranking in the race data analysis, read "Getting Started depth study of image processing technology" e-books, e-books while doing while watching the mark on the supporting code also done a test, learned a lot .
From machine learning, start with the basic concepts of image processing, deep learning step by step explains the basic principles of image processing technology and the simple implementation.
After learning theory to experiment, using a convolution neural network end-to-learn, build depth convolution neural network, using a neural network to improve circulation model, evaluation model, test model. The most critical is the model can be used in actual combat, deep learning model will be imported to the project, data type conversion functions, implementation of CAM visualization, this is what I need most.
It is really a visual and graphics, are the basis of the same!
Looking at "visual SLAM fourteen stresses" the e-book, the code is very clear! Particle Filtering, KF, EKF, Batch Optimization, Lie Group, ICP. IMU-SLAM and Semantic SLAM is the future of AR.
VO concern is the relationship between the movement of the image (image feature extraction and matching) adjacent. Rear main denoising (filtering and nonlinear optimization). The main loop detects drift over time to solve the problem (memory). Mapping is to build a map (metric map, topology map).
Now two while watching e-books, take notes, debug code, learning image processing, cv computer vision algorithms to some common, some of these methods is simple, although some more complex point, but very practical, one can learn to use, on the other hand it can also be useful for writing papers.
Collating the image recognition, computer vision aspects of electronic learning materials for your reference can learn:
https://www.yuque.com/baibinng/ctyewg/lyrsyg
Learning accumulation, combat training, making progress every day!