• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.813-826
• /
• 2022

AI-based image processing technologies in various fields have been widely studied. However, the lighter the board, the more difficult it is to reduce the weight of image processing algorithm due to a lot of computation. In this paper, we propose a method using deep learning for object recognition algorithm in lightweight embedded boards. We can determine the area using a deep neural network architecture algorithm that processes semantic segmentation with a relatively small amount of computation. After masking the area, by using more accurate deep learning algorithm we could operate object detection with improved accuracy for efficient neural network (ENet) and You Only Look Once (YOLO) toward executing object recognition in real time for lightweighted embedded boards. This research is expected to be used for autonomous driving applications, which have to be much lighter and cheaper than the existing approaches used for object recognition.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.21 no.3
• /
• pp.83-95
• /
• 2022

This paper proposes a system that performs object detection and distance estimation for application to autonomous vehicles. Object detection is performed by a network that adjusts the split grid to the input image ratio using the characteristics of the recently actively used deep learning model YOLOv4, and is trained to a custom dataset. The distance to the detected object is estimated using a bounding box and homography. As a result of the experiment, the proposed method improved in overall detection performance and processing speed close to real-time. Compared to the existing YOLOv4, the total mAP of the proposed method increased by 4.03%. The accuracy of object recognition such as pedestrians, vehicles, construction sites, and PE drums, which frequently occur when driving to the city center, has been improved. The processing speed is approximately 55 FPS. The average of the distance estimation error was 5.25m in the X coordinate and 0.97m in the Y coordinate.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.350-352
• /
• 1999

하위레벨 이미지프로세싱(Low-Level Image Processing)과 이미지인식과 해석을 주로하는 상위레벨 이미지프로세싱(High-Level Image Processing)의 접목은 현존하는 기술과 연구소서는 상대적으로 접목이 힘들며 아직까지도 많은 연구가 진행되고 있다. 후자에 더 가까운 접근을 위해서 본 논문에서는 특정 이미지를 인식하는 과정에서 모양-기반 객체(Shaped-Based Object)와 기계학습(Machine Learning) 이론을 바탕으로 두 분야의 연관을 시도하였다. 이미지 내의 객체에 대한 기하학적인 특징을 얻기 위해서 모양-기반의 특징값 추출방법을 제시하고 있으며, 보다 발전된 인식을 위해서 기계학습이론을 적용시키고 있다.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.474-474
• /
• 2000

This paper shows that a system with two-link arm can obtain hand reaching movement to a target object projected on a visual sensor by reinforcement learning using a layered neural network. The reinforcement signal, which is an only signal from the environment, is given to the system only when the hand reaches the target object. The neural network computes two joint torques from visual sensory signals, joint angles, and joint angular velocities considering the urn dynamics. It is known that the trajectory of the voluntary movement o( human hand reaching is almost straight, and the hand velocity changes like bell-shape. Although there are some exceptions, the properties of the trajectories obtained by the reinforcement learning are somewhat similar to the experimental result of the human hand reaching movement.

• Proceedings of the KIEE Conference
• /
• 2007.10a
• /
• pp.407-408
• /
• 2007

In this paper, we present a new method for classifying object using LVQ (Learning Vector Quantization) with flexible output layer. The proposed LVQ is a supervised learning method that dynamically generates output neurons and initializes automatically the weight vectors from training patterns. If the classes of the nearest output neuron is different from the class of the training pattern, a new output neuron is created and the given training pattern is used to initialize the weight vector of the created neuron. The proposed method is significantly different from the previous competitive learning algorithms in the point that the output neurons are dynamically generated during the learning process.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.68 no.2
• /
• pp.359-363
• /
• 2019

As commercial drones have been widely used, concerns for collision accidents with people and invading secured properties are emerging. The detection of drone is a challenging problem. The deep learning based object detection techniques for detecting drones have been applied, but limited to the specific cases such as detection of drones from bird and/or background. We have tried not only detection of drones, but classification of different drones with an end-to-end model. YOLOv2 is used as an object detection model. In order to supplement insufficient data by shooting drones, data augmentation from collected images is executed. Also transfer learning from ImageNet for YOLOv2 darknet framework is performed. The experimental results for drone detection with average IoU and recall are compared and analysed.

• International Journal of Advanced Culture Technology
• /
• v.10 no.1
• /
• pp.302-309
• /
• 2022

Occupational safety accidents are caused by various factors, and it is difficult to predict when and why they occur, and it is directly related to the lives of workers, so the interest in safety accidents is increasing every year. Therefore, in order to reduce safety accidents at industrial fields, workers are required to wear personal protective equipment. In this paper, we proposes a method to automatically check whether workers are wearing safety helmets among the protective equipment in the industrial field. It detects whether or not the helmet is worn using YOLOv5, a computer vision-based deep learning object detection algorithm. We transfer learning the s model among Yolov5 models with different learning rates and epochs, evaluate the performance, and select the optimal model. The selected model showed a performance of 0.959 mAP.

• The Journal of Korea Robotics Society
• /
• v.17 no.1
• /
• pp.40-47
• /
• 2022

A reward function suitable for a task is required to manipulate objects through reinforcement learning. However, it is difficult to design the reward function if the ample information of the objects cannot be obtained. In this study, a demonstration-based object manipulation algorithm called stochastic exploration guided by demonstration (SEGD) is proposed to solve the design problem of the reward function. SEGD is a reinforcement learning algorithm in which a sparse reward explorer (SRE) and an interpolated policy using demonstration (IPD) are added to soft actor-critic (SAC). SRE ensures the training of the critic of SAC by collecting prior data and IPD limits the exploration space by making SEGD's action similar to the expert's action. Through these two algorithms, the SEGD can learn only with the sparse reward of the task without designing the reward function. In order to verify the SEGD, experiments were conducted for three tasks. SEGD showed its effectiveness by showing success rates of more than 96.5% in these experiments.

• Journal of the Semiconductor & Display Technology
• /
• v.21 no.1
• /
• pp.108-113
• /
• 2022

The deep-learning-based measurement method with the through-focus scanning optical microscopy (TSOM) estimated the size of the object using the classification. However, the measurement performance of the method depends on the number of subdivided classes, and it is practically difficult to prepare data at regular intervals for training each class. We propose an approach to measure the size of an object in the TSOM image using the deep-learning regression model instead of using classification. We attempted our proposed method to estimate the top critical dimension (TCD) of through silicon via (TSV) holes with 2461 TSOM images and the results were compared with the existing method. As a result of our experiment, the average measurement error of our method was within 30 nm (1σ) which is 1/13.5 of the sampling distance of the applied microscope. Measurement errors decreased by 31% compared to the classification result. This result proves that the proposed method is more effective and practical than the classification method.

• Journal of Convergence for Information Technology
• /
• v.10 no.8
• /
• pp.40-46
• /
• 2020

As interest in traffic safety increases, research on autonomous driving, which reduces the incidence of traffic accidents, is increased. Object recognition and detection are essential for autonomous driving. Therefore, research on object recognition and detection through traffic image big data is being actively conducted to determine the road conditions. However, because most existing studies use only daytime data, it is difficult to recognize objects on night roads. Particularly, in the case of a light source object, it is difficult to use the features of the daytime as it is due to light smudging and whitening. Therefore, this study proposes Yolo based light source object detection for traffic image big data processing. The proposed method performs image processing by applying color model transitions to night traffic image. The object group is determined by extracting the characteristics of the object through image processing. It is possible to increase the recognition rate of light source object detection on a night road through a deep learning model using candidate group data.