• Current Optics and Photonics
• /
• v.2 no.6
• /
• pp.595-605
• /
• 2018

The two-photon process of microscopy provides good spatial resolution and optical sectioning ability when observing quasi-static endogenous fluorescent tissue within an in vivo animal model skin. In order to extend the use of such systems, we developed a two-photon laser scanning microscopy system capable of also capturing $512{\times}512$ pixel images at 90 frames per second. This was made possible by incorporating a 72 facet polygon mirror which was mounted on a 55 kRPM motor to enhance the fast-scan axis speed in the horizontal direction. Using the enhanced temporal resolution of our high-speed two-photon laser scanning microscope, we show that rapid processes, such as fluorescently labeled erythrocytes moving in mouse blood flow at up to 1.2 mm/s, can be achieved.

• Clinical Endoscopy
• /
• v.55 no.3
• /
• pp.390-400
• /
• 2022

Background/Aims: Previous artificial intelligence (AI) models attempting to segment gastric intestinal metaplasia (GIM) areas have failed to be deployed in real-time endoscopy due to their slow inference speeds. Here, we propose a new GIM segmentation AI model with inference speeds faster than 25 frames per second that maintains a high level of accuracy. Methods: Investigators from Chulalongkorn University obtained 802 histological-proven GIM images for AI model training. Four strategies were proposed to improve the model accuracy. First, transfer learning was employed to the public colon datasets. Second, an image preprocessing technique contrast-limited adaptive histogram equalization was employed to produce clearer GIM areas. Third, data augmentation was applied for a more robust model. Lastly, the bilateral segmentation network model was applied to segment GIM areas in real time. The results were analyzed using different validity values. Results: From the internal test, our AI model achieved an inference speed of 31.53 frames per second. GIM detection showed sensitivity, specificity, positive predictive, negative predictive, accuracy, and mean intersection over union in GIM segmentation values of 93%, 80%, 82%, 92%, 87%, and 57%, respectively. Conclusions: The bilateral segmentation network combined with transfer learning, contrast-limited adaptive histogram equalization, and data augmentation can provide high sensitivity and good accuracy for GIM detection and segmentation.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.07a
• /
• pp.269-270
• /
• 2012

Features which exhibit scale and rotation invariance, such as SIFT, are notorious for expensive computation time, and often overlooked for real-time tracking scenarios. This paper proposes a descriptorless matching algorithm based on motion vectors between consecutive frames to find the geometrically closest candidate to each tracked reference feature in the database. Descriptor-less matching forgoes expensive SIFT descriptor extraction without loss of matching accuracy and exhibits dramatic speed-up compared to traditional, naive matching based trackers. Descriptor-less SIFT tracking runs in real-time on an Intel dual core machine at an average of 24 frames per second.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.9
• /
• pp.2189-2196
• /
• 2014

In this paper, a VLIW(Very Long Instruction Word) vision coprocessor which can efficiently accelerate computer vision algorithm for automotive is designed. The VLIW coprocessor executes four instructions per clock cycle via 8-stage pipelined structure and has 36 integer and floating-point instructions to accelerate computer vision algorithm for pedestrian detection. The processor has about 300-MHz operating frequency and about 210,900 gates under 45nm CMOS technology and its estimated performance is 1.2 GOPS(Giga Operations Per Second). The vision system composed of vision primitive engine and eight VLIW coprocessors can execute pedestrian detection at 25~29 frames per second(FPS). Because the VLIW coprocessor has high detection rate and loosely coupled interface with host processor, it can be efficiently applicable to a wide range of vision applications.

• Journal of IKEEE
• /
• v.18 no.4
• /
• pp.581-585
• /
• 2014

In this paper, we propose a pedestrian detection by applying the HOG feature using ROI. Conventional HOG method has high accuracy, but shows the disadvantage of slow processing speed. By applying the ROI to the conventional method reduce computations for unnecessary area. Therefore proposed method improves the processing speed. In order to set the ROI area, we propose a structure that combined odd frames and even frames. Odd frame is in charge of operation for the entire area. And even frame does the operation for the ROI area. Implementation results of proposed method maintaining the same accuracy as the conventional method show a 20% improved performance of 8.3 frames per second.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.1
• /
• pp.133-142
• /
• 2012

In this paper, we propose and implement a high-speed algorithm for CGH that is to calculate digital hologram by modeling the interference phenomenon for tow lights. This algorithm changes the computation equations into a parallel-computable ones and implements it with a structure consisting of two kinds of cells (initial calculation cell, and update calculation cell). The parallel computation algorithm is to get the rest hologram pixels concurrently after calculation the first hologram column. Here, the initial calculation cells compute the first column of the hologram and the update calculation cells compute the rest of the hologram. The two kinds of cells performs a pipeline operation to complete the operations of the two cells at the same time. A CGH calculator to compute the hole hologram for a light source is structured by arranging the two kinds of cells. Results from simulation showed that the maximum operation frequency is about 215MHz. So, experiments are performed by setting this frequency and the same environments as the method showing the best performance. As the results, the proposed one could complete the computation of 81.75 CGH frames per second, while the previous method computes 62.9 CGH frames per second.

• Journal of KIISE
• /
• v.42 no.5
• /
• pp.566-572
• /
• 2015

We proposed a novel energy optimization technique for smartphone GPUs, more aggressively lowering the GPU frequency while obtaining higher energy efficiency with a negligible negative impact on the GPU performance. In order to achieve the Quality of Service (QoS) specified by the smartphone application, the proposed optimization technique employed the minimal acceptable GPU frequency based on average Frames per Second (FPS) for each GPU frequency level. Our experimental results on a smartphone development board showed that the proposed technique can reduce the GPU energy consumption by up to 23% over the default DVFS algorithm with only a 0.45 frame drop.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.23 no.4
• /
• pp.99-105
• /
• 2023

In this paper, an intelligent CCTV was designed and experimentally developed by using an IOT device, Raspberry Pi, and artificial intelligence technology. Object Detection technology was used to detect the number of people on the CCTV screen, and Action Detection technology provided by OpenPose was used to detect emergency situations. The proposed system has a structure of CCTV, server and client. CCTV uses Raspberry Pi and USB camera, server uses Linux, and client uses iPhone. Communication between each subsystem was implemented using the MQTT protocol. The system developed as a prototype could transmit images at 2.7 frames per second and detect emergencies from images at 0.2 frames per second.

• Proceedings of the IEEK Conference
• /
• 1999.11a
• /
• pp.1085-1088
• /
• 1999

This paper presents a method for soccer scene analysis and coordinate transformation from scene to ground model using a priori knowledge. First, the ground and spectator regions are separated, and various objects are extracted from the separated ground region. Second, an affine model is used for mapping the object positions on the soccer image into the position on the ground model. Problems regarding holes arising from mapping processing are solved using inverse mapping instead of a usual interpolation method. Experiments are performed on a PC using about 100 RGB images acquired at 240＊640 resolution and 3∼5 frames per second.

• Journal of information and communication convergence engineering
• /
• v.18 no.4
• /
• pp.267-277
• /
• 2020

It is important to understand the exact habitat distribution of endangered species because of their decreasing numbers. In this study, we build a system with a deep learning module that collects the image data of endangered animals, processes the data, and saves the data automatically. The system provides a more efficient way than human effort for classifying images and addresses two problems faced in previous studies. First, specious answers were suggested in those studies because the probability distributions of answer candidates were calculated even if the actual answer did not exist within the group. Second, when there were more than two entities in an image, only a single entity was focused on. We applied an object detection algorithm (YOLO) to resolve these problems. Our system has an average precision of 86.79%, a mean recall rate of 93.23%, and a processing speed of 13 frames per second.