Search

With the recent emergence of artificial intelligence (AI) technology, autonomous vehicle industry has rapidly adopted this technology to investigate self-driving systems based on AI technology. Although autonomous driving is frequently used in high-speed environments, most studies are conducted on low-speed driving on complex urban roads. Currently, most commercialized self-driving cars in SAE autonomous driving level 2 provide practical performance on high-speed roads using various sensors. However, these systems have to process huge sensor data and apply complex control algorithms. Recently, studies have been conducted on the use of image-based end-to-end deep learning to control autonomous driving systems that can be configured at a low cost without expensive sensors and complex processes. In this study, we proposed an autonomous driving control system using a novel end-to-end deep learning model for high-speed environments, and also compared the performance of the proposed system with NVIDIA end-to-end driving system.

This paper proposes a smart machine-learning-based unmanned aerial vehicle (UAV) control system that optimizes UAV operations in real time. The purpose of this system is to increase the efficiency of UAVs that need to operate with limited resources. This can be accomplished by allowing the UAVs in flight to identify their current state and respond appropriately. The proposed system, which is developed based on “cloud robotics,” benefits from the powerful computational capabilities of cloud computing and can therefore calculate many types of information received from various sensors in real time to maximize the performance of the UAV control system. The system learn about normal situations when creating models. That is, preprocessing data that is correlated with a particular situation and modeling it with a “multivariate Gaussian distribution.” Once the model is created, the UAV can be used to analyze the current situation in real time during flight. Of course, it is possible to recognize the situation based on the traditional RULE or the latest LSTM. However, this is not an appropriate solution for UAV situations where irregularities are severe and unpredictable. In this paper, we succeeded in recognizing the UAV flight status in real time by the proposed method and succeeded in optimizing it by adjusting communication cycle based on a recognized situation. Based on the results of this study, we expect to be able to stabilize and optimize systems that are highly irregular and unpredictable. In other words, this system will be extended to learn about various situations and create a model. A reliable and efficient smart system can be designed by judging the situation comprehensively.

This paper proposes a network model that classifies objects in images by reflecting texture and shape in a balanced way. In the case of the existing image classification model, objects are classified depending on the texture of the object. Because of this characteristic, if noise occurs in the image or the texture is intentionally changed, the image classification model is likely to misclassify the object. In addition, it can be deliberately misclassified by noise generated by hostile attack, and if it is used in industrial sites closely related to humans, a big problem may occur. To solve these problems, this paper proposes a method of extracting texture and shape features from the classification image and blending them appropriately to classify objects in the image appropriately. The results of comparing and evaluating the performance with the existing model and a new network model proposed in this paper for a general image, texture converted image, and an image with hostile attack noise added are as follows: Similar performance was obtained in the general images, but the accuracy improved by 54.64% and 69.02% for the image with textured and hostile attack noised added when using the newly proposed network model.

Object detection is a technology that deals with recognizing classes of objects and their location. It is used in many different areas, such as in face-detecting systems [16, 34, 37], surveillance tools [9], human-machine interfaces [17], and self-driving cars [18, 23, 25, 26, 30]. These days, deep learning object detection approaches have achieved significantly better performance than the classical feature-based algorithms. Darknet [31] is a deep learning object detection framework, which is well known for its fast speed and simple structure. Unfortunately, Darknet can only work with Nvidia CUDA [6] for accelerating its deep learning calculations. For this reason, users have only limited options of selecting appropriate graphic cards. Open computing language (OpenCL) [35], an open standard for cross-platform, parallel programming of heterogeneous systems, is available for the general hardware accelerators. However, many deep learning frameworks including Darknet have no support for OpenCL. In our previous paper, we presented OpenCL-Darknet [19], which transformed the CUDA-based Darknet into an open standard OpenCL backend. The original OpenCL-Darknet successfully showed its ability for the general graphics processing unit (GPU) hardware. However, it could not achieve competitive performance compared with the CUDA version, and it only supported a limited platform. In this study, we improved the performance of OpenCL-Darknet with several optimization techniques and added support for various architectures. We also evaluated OpenCL-Darknet not only in AMD R7 accelerated processing unit (APU) with OpenCL 2.0, but also in Nvidia GPU and ARM Mali embedded GPU with OpenCL 1.2 Profile. The evaluation using the standard object detection datasets showed that our advanced OpenCL-Darknet reduced the processing time by at most 50% on average for various deep learning object detection networks compared with our original implementation. We also showed that our OpenCL deep learning framework has competitiveness compared with the CUDA-based one.