• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.22 no.3
• /
• pp.452-458
• /
• 2013

Over the past several years, many studies have been carried out in the field of 3D data inspection systems. Several attempts have been made to improve the quality of manufactured parts. The introduction of laser sensors for inspection has made it possible to acquire data at a remarkably high speed. In this paper, a robust inspection technique for detecting defects in 3D pressed parts using laser-scanned data is proposed. Point cloud data are segmented for the extraction of features. These segmented features are used for shape matching during the localization process. An iterative closest point (ICP) algorithm is used for the localization of the scanned model and CAD model. To achieve a higher accuracy rate, the ICP algorithm is modified and then used for matching. To enhance the speed of the matching process, aKd-tree algorithm is used. Then, the deviation of the scanned points from the CAD model is computed.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.7 no.11
• /
• pp.2824-2838
• /
• 2013

There has been a lot of attention paid recently to analyze dynamic human gestures that vary over time. Most attention to dynamic gestures concerns with spatio-temporal features, as compared to analyzing each frame of gestures separately. For accurate dynamic gesture recognition, motion feature extraction algorithms need to find representative features that uniquely identify time-varying gestures. This paper proposes a new feature-extraction algorithm using temporal self-similarity based on a hierarchical human model. Because a conventional temporal self-similarity method computes a whole movement among the continuous frames, the conventional temporal self-similarity method cannot recognize different gestures with the same amount of movement. The proposed model-based temporal self-similarity method groups body parts of a hierarchical model into several sets and calculates movements for each set. While recognition results can depend on how the sets are made, the best way to find optimal sets is to separate frequently used body parts from less-used body parts. Then, we apply a multiclass support vector machine whose optimization algorithm is based on structural support vector machines. In this paper, the effectiveness of the proposed feature extraction algorithm is demonstrated in an application for taebo gesture recognition. We show that the model-based temporal self-similarity method can overcome the shortcomings of the conventional temporal self-similarity method and the recognition results of the model-based method are superior to that of the conventional method.

• Journal of Korean Institute of Industrial Engineers
• /
• v.19 no.2
• /
• pp.119-132
• /
• 1993

This paper deals with the development of an expert process design system interfaced with CAD for porismatic parts in injection mold manufacture. The developed CAD/CAPP system consists of two modules such as CAD interface module and process design module. Parts are represented using AutoCAD system on the IBM PC/AT. CAD interface module recognizes form features and manufacturing features of the part using form feature recognition algorithm and manufacturing feature recognition rule base. Process design module selects operations and determines machine tools, cutting tools and operation sequencing by using knowledge base which is acquired from expert process planners. A case study is implemented to evaluate feasibilities of the function of the proposed system. The CAD/CAPP system can improve the efficiency of process design activities and reduce the time required for process design.

• Transactions of the Korean Society of Machine Tool Engineers
• /
• v.15 no.6
• /
• pp.121-126
• /
• 2006

Snap-fit is being used in manufacture of plastic products. Integral features using snap-fit are classified as locks, locators and enhancements. Locking features complete the process of attachment by providing physical interference to prevent separation. Looking feature pairs consist of two components, i.e., a flexible latch and a rigid catch and require particular care and attention for their selection. We can make several locking feature pairs by selecting latch and catch, but some parts restrict freedom of selection. Therefore, part designers must know the characteristic properties of generic locking feature forms as considering a specific design problem. In this paper, it has been presented about problem of small size products using locking feature and then introduced new locking feature applicable to small parts.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 1995.04a
• /
• pp.559-564
• /
• 1995

Feature based design approach is widely studied for the application of mechanical part design and process planning. Mechanical parts are associated with volumetric form features in nature. Therfore, one of the important characteristics that reside in the form feature research until now is that features have been studied in connection with CAPP for material removal. We studied the application of feature based design for ship structure design. Ship structure has interesting nature that tis distinct from mechanical parts. Among these are multiple cell structure, non-volumetric part and production by welding or assembling. An idea of applying feature based design paradigm for design, process planning, cost analysis and engineering calculation was shown. Non-manifold geometric modeler ACIS was adopted to fully benefit from the non-manifold nature of ship structure.

• Journal of Korean Institute of Industrial Engineers
• /
• v.21 no.1
• /
• pp.51-65
• /
• 1995

This paper presents an automatic fixture planning system for machining processes of prismatic parts. A rationalized approach to integrate fixture planning with process planning is proposed and representation schemes for workpiece, part design information with features, machine tools, cutting tools and fixtures are developed. The proposed system implements two activities of fixture planning such as machining of reference surfaces and machining of features. For machining of reference surfaces, the machining sequence of reference surfaces is determined by using decision tables, which are drawn from relations of part dimension, degree of surface roughness, fixture type and its capacity, cutting tool's capacity and experienced planners' knowledge. For machining of features, a preferential machining orientation is selected for its feature which can be machined in more than one direction, and features with the same machining orientation are grouped, and the machining sequence of features is determined by interactive mode. A case study is performed to show the performance of the proposed system.

• Management Science and Financial Engineering
• /
• v.19 no.1
• /
• pp.1-10
• /
• 2013

Hierarchical forecasting strategy does not always outperform direct forecasting strategy. The performance generally depends on demand features. This research guides the use of the alternative forecasting strategies according to demand features. This paper developed and evaluated various classification models such as logistic regression (LR), artificial neural networks (ANN), decision trees (DT), boosted trees (BT), and random forests (RF) for predicting the relative performance of the alternative forecasting strategies for the South Korean navy's spare parts demand which has non-normal characteristics. ANN minimized classification errors and inventory costs, whereas LR minimized the Brier scores and the sum of forecasting errors.

• Journal of Computational Design and Engineering
• /
• v.1 no.1
• /
• pp.48-54
• /
• 2014

In this paper, the authors propose a system for assisting mold designers of plastic parts. With a CAD model of a part, the system automatically determines the optimal ejecting direction of the part with minimum undercuts. Since plastic parts are generally very thin, many rib features are placed on the inner side of the part to give sufficient structural strength. Our system extracts the rib features from the CAD model of the part, and determines the possible ejecting directions based on the geometric properties of the features. The system then selects the optimal direction with minimum undercuts. Possible ejecting directions are represented as discrete points on a Gauss map. Our new point distribution method for the Gauss map is based on the concept of the architectural geodesic dome. A hierarchical structure is also introduced in the point distribution, with a higher level "rough" Gauss map with rather sparse point distribution and another lower level "fine" Gauss map with much denser point distribution. A system is implemented and computational experiments are performed. Our system requires less than 10 seconds to determine the optimal ejecting direction of a CAD model with more than 1 million polygons.

• Korean Journal of Computational Design and Engineering
• /
• v.5 no.2
• /
• pp.196-205
• /
• 2000

This paper proposes a topology-based algorithm for recognizing the passage features using a concept of multi-face hole loop. The Multi-face hole loop is a concetpual hole loop that is formed over several connected faces. A passage feature is recognized in the proposed approach by two multi-face hole loops that constitute its enterance and exit. The algorithm proposed in this paper checks the connectivity of the two multi-face hole loops to recognize passage features. The total number of passage features in a part is calculated from Euler equation and is compared with the number of found passage features to decide when to terminate. To find all multi-face hole loops in a part, this paper proposes an algorithm for finding all combinations of connected faces. The edge convexity is used to judge the validity of multi-face hole loops. By using the algorithm proposed in this paper, the passage features could be recognized effectively. The approach proposed in this paper is illustrated with several example parts.