Introduction

Introduction

 Sunghoon Hong (Ph.D. Student) received the M.S. degree in Intelligent Robot Engineering from the Hanyang University, Seoul, Korea, in 2016. His main interest is Human-Like Autonomous Driving Systems. He has a lot of experience in autonomous driving technologies such as Simultaneous Localization And Mapping (SLAM), Advanced Driver Assistance Systems (ADAS), Proportional-Integral-Differential (PID) control, machine learning, path-planning and navigation algorithms and has published several journal/conference papers. He is currently pursuing toward obtaining his Ph.D. degree in the Department of Electronic and Electrical Engineering, Kyungpook National University, Daegu, Korea from 2021. He is researching technologies to optimize deep learning-based object detection algorithms for human-like artificial intelligence autonomous driving systems to be applied to low-power embedded systems.

Publications

Journal Publications (KCI 2, SCI 1)

International Conference Publications (Intl. 4)

Patents (Domestic 1)

Domestic Conference Publications

Research Topic

 Eyes are the most important factor for a person to drive a vehicle. Human-like autonomous driving systems use camera sensors to recognize vehicles, bicycles, people, lanes, roads, signs, traffic lights, etc. through deep learning technology. This information helps estimate the vehicle's current location and drive autonomously along its global path to its destination. However, there is a limit to directly applying deep learning technology to embedded systems, algorithm optimization techniques are required. In this way, I would like to intensively research optimization techniques to develop vision-based human-like autonomous driving technologies in real-time low-power embedded systems.

 Forward vehicle detection is the key technique to preventing car incident in front. Artificial intelligence (AI) techniques are used to more accurately detect vehicles, but AI-based vehicle detection takes a lot of processing time due to its high computational complexity. When there is a risk of collision with a vehicle in front, the slow detection speed of the vehicle may lead to an accident. To quickly detect a vehicle in real-time, a high-speed and lightweight vehicle detection technique with similar detection performance to that of an existing AI-based vehicle detection is required. Also, to apply forward collision warning system (FCWS) technology to vehicles, it is important to provide high performance based on low-power embedded systems because the vehicle’s battery consumption must remain low. The vehicle detection algorithm occupies the most resources in FCWS. To reduce power consumption, it is important to reduce the computational complexity of an algorithm, that is, the amount of resources required to run it. This paper describes a method for fast, accurate forward vehicle detection using machine learning and deep learning. To detect a vehicle in consecutive images consistently, a Kalman filter is used to predict the bounding box based on the tracking algorithm and correct it based on the detection algorithm. As a result, its vehicle detection speed is about 25.85 times faster than deep-learning-based object detection is, and its detection accuracy is better than machine-learning-based object detection is.

 Convolutional neural networks with powerful visual image analysis of deep structures are gaining popularity in many research fields. The main difference in convolutional neural networks compared to other artificial neural networks is the addition of many convolutional layers. The convolutional layer improves the performance of artificial neural networks by extracting feature maps required for image classification. However, for applications that require very low-latency on limited processing resources, the success of a convolutional neural network depends on how fast we can compute. In this paper, we propose a novel convolution technique of fast algorithms for convolutional neural networks using continuous differential images. The proposed method improves the response speed of the algorithm by reducing the computational complexity of the convolutional layer. It is compatible with all types of convolutional neural networks, and the lower the difference in the continuous images, the better the performance. We use the darknet network to benchmark the CPU implementation of our algorithm and show state-of-the-art throughput at pixel difference thresholds from 0 to 25 pixels.

 Convolutional neural networks with powerful visual image analysis for artificial intelligence are gaining popularity in many research fields, leading to the development of various high-performance algorithms for convolution operators present in these networks. One of these approaches is implemented with general matrix multiplication (GEMM) using the well-known im2col transform for fast convolution operations. In this paper, we propose a multi-core processor-based convolution technique for high-speed convolutional neural networks (CNNs) using differential images. The proposed method improves the convolutional layer's response speed by reducing the computational complexity and using multi-thread technology. In addition, the proposed algorithm has the advantage of being compatible with all types of CNNs. We use the darknet network to evaluate the convolutional layer's performance and show the best performance of the proposed algorithm when using 4-thread parallel processing.

Projects

Contacts

Email: hopsison@gmail.com

Phone: +82 10-5241-1763

Office:

Lab: AISoC Lab.