• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.1
• /
• pp.141-144
• /
• 2020

Convolutional neural networks (CNNs) show high performance in computer vision tasks including object detection, but a lot of weight storage and computation is required. In this paper, a pruning scheme is applied to CNNs for object detection, which can remove much amount of weights with a negligible performance degradation. Contrary to the previous ones, the pruning scheme applied in this paper considers the base accelerator architecture. With the consideration, the pruned CNNs can be efficiently performed on an ASIC or FPGA accelerator. Even with the constrained pruning, the resulting CNN shows a negligible degradation of detection performance, less-than-1% point degradation of mAP on VOD0712 test set. With the proposed scheme, CNNs can be applied to objection dtection efficiently.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.562-565
• /
• 2022

Recently, the increasing number of overloaded vehicles on the road poses a risk to traffic safety, such as falling objects, road damage, and chain collisions due to the abnormal weight distribution, and can cause great damage once an accident occurs. However, this irregular weight distribution is not possible to be recognized with the current weight measurement system for vehicles on roads. To address this limitation, we propose to build an object detection-based AI model to identify overloaded vehicles that cause such social problems. In addition, we present a simple yet effective method to construct an object detection model for the large-scale vehicle images. In particular, we utilize the large-scale of vehicle image sets provided by open AI-Hub, which include the overloaded vehicles from the CCTV, black box, and hand-held camera point of view. We inspected the specific features of sizes of vehicles and types of image sources, and pre-processed these images to train a deep learning-based object detection model. Finally, we demonstrated that the detection performance of the overloaded vehicle was improved by about 23% compared to the one using raw data. From the result, we believe that public big data can be utilized more efficiently and applied to the development of an object detection-based overloaded vehicle detection model.

• Journal of the Korean Society for Precision Engineering
• /
• v.19 no.4
• /
• pp.109-116
• /
• 2002

The solution of the aircraft wing spar cross-section area optimization problem is obtained by the response surface method. The object function of the problem is wing total weight, design variables are spar cross-section areas, constraints are the conditions that the stresses at the each spar is less than the allowable stress. D-Optimal condition is utilized to obtain the experimental points to construct the response surfaces. D-Optimal experimental points are obtained by the commercial software "Deign-Expert". Response values for the object function and constraints for each experimental point are calculated by the NASTRAN. Response surfaces for object function and constraints are approximated from the response values by the least square method. The optimization solution is obtained by the DOT for the response surfaces of object function and constraints. The optimization results obtained from the response surface are compared with the results obtained by the NASTRAN SOL200.

• Proceedings of the KIEE Conference
• /
• 1998.07b
• /
• pp.490-492
• /
• 1998

An intention of this paper is design of a gyro actuator for the attitude control of an unstructured object. It is well known that the attitude control of an object hanging with wire is not easy using usual actuators. Even though an actuator such as a pan can be used for control of the object, it is difficult to meet a desired control objectives. We, for this reason, propose a gyro actuator for the attitude control of an unstructured object. The proposed gyro actuator consists of two motors. The first motor is responsible to spin the wheel and the second motor is used to turn the outer gimbal. Appling the torque to the second motor, which results in the turn of the outer gimbal, torque about the vertical axis will be obtained while a wheel of the gyro is spinning constantly. This torque is used to control the attitude of the object attached. The aim of this paper is of deriving the transfer function of the actuator and presenting the guideline of the design parameters such as the weight and the dimension of the wheel, motors, and the load capacity. Simulations to the mathematical model which has a state feedback control are conducted to show the validity of the proposed gyro actuator.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.12
• /
• pp.6009-6027
• /
• 2019

Support vector machine (SVM) classifiers have been widely used for object detection. These methods usually locate the object by finding the region with maximal score in an image. With bag-of-features representation, the SVM score of an image region can be written as the sum of its inside feature-weights. As a result, the searching process can be executed efficiently by using strategies such as branch-and-bound. However, the feature-weight derived by optimizing region classification cannot really reveal the category knowledge of a feature-point, which could cause bad localization. In this paper, we represent a region in an image by a collection of local feature-points and determine the object by the region with the maximum posterior probability of belonging to the object class. Based on the Bayes' theorem and Naive-Bayes assumptions, the posterior probability is reformulated as the sum of feature-scores. The feature-score is manifested in the form of the logarithm of a probability ratio. Instead of estimating the numerator and denominator probabilities separately, we readily employ the density ratio estimation techniques directly, and overcome the above limitation. Experiments on a car dataset and PASCAL VOC 2007 dataset validated the effectiveness of our method compared to the baselines. In addition, the performance can be further improved by taking advantage of the recently developed deep convolutional neural network features.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.1
• /
• pp.364-380
• /
• 2016

The most popular approach in object classification is based on the bag of visual-words model, which has several fundamental problems that restricting the performance of this method, such as low time efficiency, the synonym and polysemy of visual words, and the lack of spatial information between visual words. In view of this, an object classification based on weakly supervised E2LSH and saliency map weighting is proposed. Firstly, E2LSH (Exact Euclidean Locality Sensitive Hashing) is employed to generate a group of weakly randomized visual dictionary by clustering SIFT features of the training dataset, and the selecting process of hash functions is effectively supervised inspired by the random forest ideas to reduce the randomcity of E2LSH. Secondly, graph-based visual saliency (GBVS) algorithm is applied to detect the saliency map of different images and weight the visual words according to the saliency prior. Finally, saliency map weighted visual language model is carried out to accomplish object classification. Experimental results datasets of Pascal 2007 and Caltech-256 indicate that the distinguishability of objects is effectively improved and our method is superior to the state-of-the-art object classification methods.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 1999.05a
• /
• pp.224-229
• /
• 1999

In distributed object oriented environment based upon wide heterogeneous network, effective cooperation policies between/among distributed objects are needed to resolve a complexity of management of distributed objects because of growing of a large stale of systems. Thus, in this paper, we propose three trading cooperation models between/among traders for supporting a high speed and a wide selection of trader service for clients, by considering three different cooperation models(light weight trader, simple negotiation and federation) depending upon their facilities, goals, and weights of goals.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.1934-1937
• /
• 2005

This paper proposes suitable gait generation for dynamic walking of biped robot with varying load in real time. Author proposes the relationship between ZMP(Zero Moment Point) and measurement from FSR(Force Sensing Register). Simplifying this relationship, it is possible to reduce the computational time and control the biped robot in real time. If the weight of the biped robot varies in order to move some object, then joint trajectories of the the biped robot must be changed. When some object is loaded on the biped robot in it's home position, FSRs can measure the variation of weight. Evaluating the relations between varying load and stable gait of the biped robot, it can walk adaptively. This relation enables the biped robot to walk properly with varying load. The simulation is also represented in this paper which shows proposed relationships.

• 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.

• Journal of Industrial Technology
• /
• v.16
• /
• pp.71-81
• /
• 1996

This paper discusses a physiological approach motivated by the study of human hands for robot hand force control. It begins with an analysis of the human's grasping behavior to see how humans determine the grasp forces. The human controls the grasp force by sensing the friction force, that is, the weight of the object which is felt on his hand, but when slip is detected by sensing skin acceleration, the grasp force becomes much greater than the minimum force required for grasping by adding the force which is proportional to the acceleration. And two methods that can predict when and how fingers will slip upon a grasped object are considered. To emulate the human's capabilities, we propose a method for determination of as grasp force, which uses the change in the friction force. Experimental results show that the proposed method can be applied to control of robot hands to grasp objects of arbitrary weight stably without skin-like slip sensors.