In the past few years, artificial intelligence has ushered in the third climax, which is due to the improvement of hardware computing power, so that deep learning algorithm can be applied on a large scale. Before the large-scale use of deep learning, image feature engineering is more dependent on the experience of engineers, and it can't get good results in the face of random and complex environment. Through a large number of manual features, through SVM training to classify, detect, segment images. With the success of Alexnet , people began to realize that feature engineering can rely on neural network to extract itself. It has been proved that the neural network trained by big data has very strong robustness to image feature extraction.This article is divided into four chapters. Chapter 1 mainly introduces the experimental significance and motivation of this article, and introduces the role of deep learning in computer vision. Chapter 2 introduces the application of deep learning algorithms, including image classification (AlexNet, VGGNet, ResNet), target detection (FasterRCNN, YOLO, SSD), video analysis (LSTM based,3DCNN), and introduces some of the most advanced algorithms The structure and their experimental results, the experimental results of the new structure can reach 790 FPS on the Tesla P100. Chapter 3 mainly introduces the achievements I made this year and the algorithms designed for effective video understanding. Chapter 4 is a summary of the entire process and points out what improvements can be made in the future.
|Date of Award||25 Aug 2020|
- University Of Strathclyde
|Sponsors||University of Strathclyde|
|Supervisor||John Soraghan (Supervisor) & Gaetano Di Caterina (Supervisor)|