Process and Implementation: Optical Flow

Process and Implementation of Optical Flow First, it is necessary to consider the calculation method of optical flow. From the explanation of the thesis, they seem to be using the Lucas-Kanade method. This method assumes that the optical flow is constant within a small window. This is a widely used method for calculating optical flow and can be found in several libraries. The library I am using was created by Piotr Dollar of the University of California, San Diego (1). The toolbox also implements a cross-correlation method to calculate optical flow as well as horn and shunk methods.

Optical flow: Use OpenCV to calculate a high density optical stream, convert the 2-channel optical flow vector (u, v) to size and direction, and save it as an RGB image for compression. Using these Resnet 3D architectures in these frames, we got 42.65% (top 1) and 68.09% (top 5) precision. These results are not comparable to state-of-the-art optical flow models, but due to time constraints during the game it is not possible to further investigate this clue. However, finding the best network architecture is equally important to find the correct optical flow pipeline.

Based on the research of this thesis, motion and optical flow were estimated using flow network. In this model, you predict where in the frame a feature will appear by predicting the movement of pixels between frames. The streaming network uses a convolution neural network that can solve the optical flow estimation problem as a supervised learning task. The paper states that their approach effectively manages computational costs efficiently while maintaining accuracy. Implement 77.8 and 15.2 fps mAP in Nvidia Tesla K40 GPU. Compared to the award-winning work of ImageNet VID Challenge 2017, Microsoft's model increased mAP by 1% while sacrificing 0.2 fps.

How Microsoft performs video object detection - the best technology in a unified video object detection architecture with a single model

Pulsed laser holography has been used in wind tunnel experiments. Optical interferometers (devices for detecting small changes in interference fringes, in this case caused by changes in air density) are usually used to study high velocity airflow around aerodynamic objects. This type of instrument is difficult to adjust and is difficult to maintain stability. In addition, all optical components (mirrors, plates, etc.) in the optical path must be high quality and robust enough to minimize distortion at high air velocity. However, holographic systems avoid the stringent requirements of optical interferometry. Record the interference refractive index change of the gas flow caused by the pressure change as the gas bends around the aerodynamic object.