• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Journal of dental hygiene science
• /
• v.17 no.3
• /
• pp.209-217
• /
• 2017

This study investigated the relationships between experiencing verbal violence and the emotional responses and coping behaviors of dental hygienists who worked in Jeonju between February 24 and March 18, 2017 to prepare strategies for managing verbal violence and establish a healthy working environment for dental hygienists. The following findings were obtained in this study. The dental hygienists primarily experienced verbal violence from patients and guardians (1.67 points). The most common verbal violence type was, "Someone treated me impolitely." The most common emotional response to verbal violence was "anger" (3.52 points). The coping style of most dental hygienists was problem-focused coping (3.28 points), followed by emotion-focused coping (2.75 points). Most hygienists with the problem-focused coping style stated that they resolve the problem through dialogue (3.51 points), while most hygienists with the emotion-focused coping style responded that they just ignore the situation (3.78 points). The relationship between experiencing verbal violence and the emotional responses and coping behaviors of dental hygienists showed a statistically significant positive correlation (p<0.05) with emotional responses and problem- and emotion-focused coping associated with experiencing verbal violence from dentists, patients, and guardians. These findings confirm the need to reduce verbal violence, control emotional responses after exposure to verbal violence, and use more problem-focused coping measures. Dental hygienists must develop interpersonal skills and communication techniques and promote professionalism in their workplace to protect themselves from verbal violence at work.

• Agricultural and Biosystems Engineering
• /
• v.4 no.1
• /
• pp.16-21
• /
• 2003

In the agricultural field, a machine vision system has been widely used to automate most inspection processes especially in quality grading. Though machine vision system was very effective in quantifying geometrical quality factors, it had a deficiency in quantifying color information. This study was conducted to evaluate color of beef using machine vision system. Though measuring color of a beef using machine vision system had an advantage of covering whole lean tissue area at a time compared to a colorimeter, it revealed the problem of sensitivity depending on the system components such as types of camera, lighting conditions, and so on. The effect of color balancing control of a camera was investigated and multi-layer BP neural network based color calibration process was developed. Color calibration network model was trained using reference color patches and showed the high correlation with L*a*b* coordinates of a colorimeter. The proposed calibration process showed the successful adaptability to various measurement environments such as different types of cameras and light sources. Compared results with the proposed calibration process and MLR based calibration were also presented. Color calibration network was also successfully applied to measure the color of the beef. However, it was suggested that reflectance properties of reference materials for calibration and test materials should be considered to achieve more accurate color measurement.

• Journal of Sensor Science and Technology
• /
• v.28 no.2
• /
• pp.94-100
• /
• 2019

Drivable area detection is a major task in advanced driver assistance systems. For drivable area detection, several studies have proposed vision-sensor-based approaches. However, conventional drivable area detection methods that use vision sensors are not suitable for environments with changes in road elevation. In addition, if the boundary between the road and vegetation is not clear, judging a vegetation area as a drivable area becomes a problem. Therefore, this study proposes an accurate method of detecting drivable areas in environments in which road elevations change and vegetation exists. Experimental results show that when compared to the conventional method, the proposed method improves the average accuracy and recall of drivable area detection on the KITTI vision benchmark suite by 3.42%p and 8.37%p, respectively. In addition, when the proposed vegetation area removal method is applied, the average accuracy and recall are further improved by 6.43%p and 9.68%p, respectively.

• Journal of Information Processing Systems
• /
• v.19 no.6
• /
• pp.730-744
• /
• 2023

The topic of this paper is the recognition of human activities using egocentric vision, particularly captured by body-worn cameras, which could be helpful for video surveillance, automatic search and video indexing. This being the case, it could also be helpful in assistance to elderly and frail persons for revolutionizing and improving their lives. The process throws up the task of human activities recognition remaining problematic, because of the important variations, where it is realized through the use of an external device, similar to a robot, as a personal assistant. The inferred information is used both online to assist the person, and offline to support the personal assistant. With our proposed method being robust against the various factors of variability problem in action executions, the major purpose of this paper is to perform an efficient and simple recognition method from egocentric camera data only using convolutional neural network and deep learning. In terms of accuracy improvement, simulation results outperform the current state of the art by a significant margin of 61% when using egocentric camera data only, more than 44% when using egocentric camera and several stationary cameras data and more than 12% when using both inertial measurement unit (IMU) and egocentric camera data.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.10 no.2
• /
• pp.67-80
• /
• 2006

In this paper, we propose a new pointing device that is replaced a mouse as the pointing device with. For reducing the existing pointing device's problem which had marker and high-cost, we develop a new pointing device using computer vision like as a similar human vision system. The proposed system first carries out a real-time movement tracking system using image data which are segmented by color modeling, and finally does the pointing action by 3-D coordinate calculated from stereo geometry information resulting from stereo matching of the segmented region.

• Journal of Biosystems Engineering
• /
• v.25 no.1
• /
• pp.55-62
• /
• 2000

Autonomous farm operation needs to be developed for safety, labor shortage problem, health etc. In this research, an autonomous tractor for tillage was investigated using machine vision and a fuzzy logic controller(FLC). Tractor heading and offset were determined by image processing and a geomagnetic sensor. The FLC took the tractor heading and offset as inputs and generated the steering angle for tractor guidance as output. A color CCD camera was used fro the image processing . The heading and offset were obtained using Hough transform of the G-value color images. 15 fuzzy rules were used for inferencing the tractor steering angle. The tractor was tested in the file and it was proved that the tillage operation could be done autonomously within 20 cm deviation with the machine vision and the FLC.

• Journal of Sensor Science and Technology
• /
• v.14 no.3
• /
• pp.191-197
• /
• 2005

When fabricating a vision chip, we should consider the noise problem, such as the fixed pattern noise(FPN) due to the process variation. In this paper, we propose an edge-detection circuit based on biological retina using the offset-free column readout circuit to reduce the FPN occurring in the photo-detector. The offset-free column readout circuit consists of one source follower, one capacitor and five transmission gates. As a result, it is simpler and smaller than a general correlated double sampling(CDS) circuit. A vision chip for edge detection has been designed and fabricated using $0.35\;{\mu}m$ 2-poly 4-metal CMOS technology, and its output characteristics have been investigated.

• Journal of Korea Water Resources Association
• /
• v.54 no.7
• /
• pp.475-484
• /
• 2021

The optimal design for water distribution system (WDS) is not only satisfying the minimum required water pressure of the nodes, but also minimizing pipe cost, etc. The number of designs of WDS increases exponentially due to the arrangement of various pipes. Various optimization algorithms were applied to propose an optimized design of WDS. In this study, Modified Hybrid Vision Correction Algorithm (MHVCA) with improved self-adapting parameter was applied to optimal design of WDS. The performance was improved by changing the Hybrid Rate (HR) of the existing Hybrid Vision Correction Algorithm (HVCA) to nonlinear HR. To verify the performance of the proposed MHVCA, it applied to mathematical problems consisting of 2 and 30 decision variables and constrained mathematical problems. In order to review the application results of MHVCA, it was compared with Harmony Search (HS), Improved Harmony Search (IHS), Vision Correction Algorithm (VCA) and HVCA. Finally, MHVCA was applied to the optimal design problem of WDS and the results were compared with other algorithms. MHVCA showed better results than other algorithms in mathematical problems and WDS problem. MHVCA will be able to show good results by applying to various water resource engineering problems as well as problems applied in this study.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.49 no.9
• /
• pp.353-361
• /
• 2012

In this paper, we propose a improvement method of vision inspector for T type welding nut using an auto sorting system. We used edge and thread detection with histogram of image which is captured by machine vision camera. We also used a binary morphology operation for a detection of spot. A major problem in this vision inspector is abnormal operation caused by degradation of image acquired. These degradations caused by oil pollution on conveyer belt. For overcome this problem, we introduce a pre-processing using a masked histogram equalization on the image acquired. Histogram equalization is applied on masked region (nut part) for increase contrast. As a result, we can remove features caused by oil pollution on background and reduce a ratio of abnormal operation from 10.0 % to 0.2 %.