Mobile QR Code QR CODE

Color Barcode Detection of Complementary Color Barcode-based Optical Camera Communications with a Deep Neural Network

(Min Tae Kim) ; (Byung Wook Kim)

Complementary color barcode-based optical camera communications (CCB-OCC), which uses electronic displays and cameras as data transmitters and receivers, can offer a simple and short-range communications interface. In this paper, we enhance the data rate of a CCB-OCC scheme by using deep neural network (DNN)-based color barcode detection and adaptive colorvalue extraction. Unlike a conventional scheme, the proposed method extracts a color barcode region using a You Only Look Once (YOLO) real-time detection model that provides reliable barcode region extraction from various viewing angles between the display and the camera. In addition, the proposed scheme extracts accurate color values from color histograms using adaptive peak positions that express consecutive signals between packets. Experimental results verify data rate improvement in the proposed CCB-OCC scheme using DNN-based barcode detection and adaptive color-value extraction in a display-to-camera (D2C) communications link.

Customer Service Assist System based on Natural Language Processing

(Nayoung Yun) ; (Sangkyu Lim) ; (Seoyoung Hong) ; (Jiwon Moon) ; (Hakjun Lee) ; (Sunmok Kim) ; (Heung-Jae Lee) ; (Ki-Baek Lee)

This paper proposes a novel assist system for customer service representatives based on natural language processing (NLP). In the proposed system, an NLP model calculates the relationships between a question from a customer and all the questions in a given FAQ list. Based on the model’s calculation, the system will recommend several FAQs that are more similar to the customer’s question than the others in the FAQ list and then the representative responses, whether the recommended questions are actually similar to the customer’s question or not. Since these responses become the data for the NLP model’s next training, the NLP model’s accuracy can be incrementally enhanced by repetitive fine-tuning with the accumulated data. The experimental result shows that the proposed system can effectively help customer service representatives as well as incrementally improve via the automatically accumulated data.

Efficient Object Detection Acceleration Methods for Autonomous-driving Embedded Platforms

(Jiwoong Choi) ; (Dayoung Chun) ; (Hyuk-Jae Lee) ; (Hyun Kim)

Object detection in autonomous vehicles is typically operated in an embedded system to reduce power consumption. The use of an object detection algorithm with high accuracy and realtime detection speed in the embedded systems is essential for ensuring safe driving. This study proposes a parallel processing method for GPU and CPU operations to enhance the detection speed of the model. In addition, this study proposes data augmentation and image resize techniques that consider the camera input size of autonomous driving, which increases the accuracy significantly while improving the detection speed. The application of these proposed schemes to a baseline algorithm, tiny Gaussian YOLOv3, improves the mean average precision by 1.14 percent points (pp) for the Berkeley Deep Drive (BDD) dataset and 1.34 pp for the KITTI dataset compared to the baseline. Furthermore, in the NVIDIA Jetson AGX Xavier, which is an embedded platform for autonomous driving, the proposed algorithm improves the detection speed by 22.54 % for the BDD, and 24.67 % for the KITTI compared to the baseline, thereby enabling high-speed real-time detection on both datasets.

A Real-time Implementation of Object Tracking Algorithms Fusion for Thermal Imaging Sequences

(Nguyen Ngoc Hung) ; (Cao Huu Tinh) ; (Nguyen Vi Thuan) ; (Pham Ngoc Van)

Object tracking is a widely used algorithm in image processing. When tracking objects on thermal images, however, issues, such as changes in size, temporary occlusion, lack of prominent features, and active thermal noise, are frequently encountered. This article proposes using Multi- Processor System on Chip (MPSoC) technology to implement new tracking algorithms on programmable hardware platforms to increase the computational speed. This real-time algorithm will combine the superiority of existing object tracking algorithms, such as Camshift, Kalman, and Optical Flow, to overcome the difficulties mentioned above.

Event-based White Blood Cell Classification from Nailfold Capillaries

(Yuli Sun Hariyani) ; (Cheolsoo Park)

Detecting absorption gaps in a nailfold capillary can be used to quantify an estimation of white blood cells (WBCs). Previously, the absorption gaps in a nailfold capillary were usually measured using a standard camera on a fingernail. However, difficulties arise due to low visibility of the gap, the small size of the capillaries, the high speed of WBC movement, and the lack of contrast between the capillary and its environment/background. To address these issues, an eventbased WBC image is utilized as input data to detect WBC existence in the nailfold. Specifically, we utilize a dynamic vision sensor (DVS) camera, which can detect a change in luminance on a pixel basis and can produce a stream of asynchronous event output at a microsecond temporal resolution. With the event-based WBC dataset, we conduct a classification task using three different machine learning algorithms: k-nearest neighbors, the decision tree, and random forest. The best result is from random forest with 75.51% accuracy. Based on our evaluation, event-based WBC classification is a promising new approach to detecting WBC presence in nailfold capillaries.

A Study on Parallel Clustering Algorithms based on MapReduce

(Garvit Chugh) ; (Nitesh Singh Bhati) ; (Puneet Kumar) ; (Vishal Bharti)

Since the advent of the global computerized market, the volume of digital information has grown exponentially, as has the demand for storing it. As the price of storage devices decreases, the necessity to analyze vast quantities of unstructured digital data to retain only essential information increases. MapReduce is a programming paradigm for producing and generating massive information indices. Using MapReduce to produce meaningful clusters from such a massive amount of raw data is an efficient way to manage such voluminous amounts of data. On the other hand, the existing industry standard for data clustering algorithms presents significant obstacles. The conventional clustering calculation efficiently handles a great deal of information from various sources, such as online media, business, and the web. Nevertheless, the sequential count in clustering approaches is time-intensive in these conventional calculations. The wide varieties of K-Means, including K-Harmonic Means, are sensitive to forming cluster centers in huge datasets. This work suggests a logical evaluation of such calculations. It offers a study of the various k-means clustering algorithms employed in MapReduce, as well as the study on the introduction and the open challenges of parallelism in MapReduce.

Consensus Building using Deep Reinforcement Learning for Energy Management

(Yuya Tarutani) ; (Isato Oishi) ; (Yukinobu Fukushima) ; (Tokumi Yokohira)

A variety of information is collected from IoT devices. As those devices become more familiar to users, network services must consider the influence of the user. We propose a method to maximize the value from power consumption and minimize the cost incurred to ensure user satisfaction. However, one problem is that user satisfaction cannot increase because it is considered a constraint on power consumption. In this paper, we propose a consensus building method to minimize power consumption and maximize user satisfaction. An exhaustive search incurs a large calculation overhead to determine device parameters. Thus, the proposed method uses reinforcement learning to solve this problem. From its evaluation, we clarify that the proposed method attains about 1.5 times the total reward compared with the conventional method. Moreover, we also clarify that 99.9% of the total reward can be achieved, compared to the exhaustive search.

Adaptive Guard Time Control for Distributed TDMA Networks with a Large Propagation Delay

(Youngeun Cho) ; (Hoki Baek)

In TDMA networks with a large propagation delay, a large guard time is required to ensure broadcasting without collisions. However, this causes considerable waste, and degrades network performance. In this paper, we propose Adaptive Guard Time with Inter-transmitter Propagation Delay (AGT-IPD) to reduce the guard time while ensuring broadcasting. Each transmitter adjusts the guard time by using the propagation delay for the next time slot owner. AGT-IPD employs a distributed mechanism for propagation delay estimation and time synchronization. We then conduct a trade-off analysis, and demonstrate efficient guard time management and overhead control from AGT-IPD from simulation modeling. Through numerical results, AGT-IPD shows improvements that provide higher link utilization than other schemes to which it is compared, regardless of packet length, maximum communication range, the number of nodes, and the number of errors.

A Fast (7, 3)-adder Circuit for High-speed Multiplier Design

(Myungchul Yoon)

A set of fast (m, 3)-adder circuits (4 ≤ m ≤ 7) is presented in this paper. An (m, 3)-adder adds m bits at a time and produces three outputs (N, C, S). These adders are designed to implement high-speed (7, 3)-adder based multipliers that use a (7, 3)-adder as a basic unit and five other adders ((6, 3), (5, 3), (4, 3), (3, 2), and (2, 2)-adder) as auxiliary units for the addition of partial products. Multipliers require adding tens of partial products to obtain a result, and they can be added by (7, 3)-adders more quickly than by (3, 2)-adders. In simulation results, the worst-case delay and power of the new (7, 3)-adder are 1.45 times and 2.4 times larger than those of the reference (3, 2)-adder. However, the parallel addition of partial products with the (7, 3)-adder is faster and consumes less power than with the (3, 2)-adder because the (7, 3)-adder based multiplier requires fewer adders and addition stages than a (3, 2)-adder based multiplier. This result shows that the speed of a multiplier can be increased by using (7, 3)-adders instead of (3, 2)-adders in reducing partial products.