• International conference on construction engineering and project management
• /
• 2022.06a
• /
• pp.877-885
• /
• 2022

Vision-based activity recognition has been widely attempted at construction sites to estimate productivity and enhance workers' health and safety. Previous studies have focused on extracting an individual worker's postural information from sequential image frames for activity recognition. However, various trades of workers perform different tasks with similar postural patterns, which degrades the performance of activity recognition based on postural information. To this end, this research exploited a concept of human-object interaction, the interaction between a worker and their surrounding objects, considering the fact that trade workers interact with a specific object (e.g., working tools or construction materials) relevant to their trades. This research developed an approach to understand the context from sequential image frames based on four features: posture, object, spatial features, and temporal feature. Both posture and object features were used to analyze the interaction between the worker and the target object, and the other two features were used to detect movements from the entire region of image frames in both temporal and spatial domains. The developed approach used convolutional neural networks (CNN) for feature extractors and activity classifiers and long short-term memory (LSTM) was also used as an activity classifier. The developed approach provided an average accuracy of 85.96% for classifying 12 target construction tasks performed by two trades of workers, which was higher than two benchmark models. This experimental result indicated that integrating a concept of the human-object interaction offers great benefits in activity recognition when various trade workers coexist in a scene.

• Journal of Korea Multimedia Society
• /
• v.18 no.1
• /
• pp.9-16
• /
• 2015

Various approaches to process active sonar signals are under study, but there are many problems to be considered. The sonar signals are distorted by the underwater environment, and the spatio-temporal and spectral characteristics of active sonar signals change in accordance with the aspect of the target even though they come from the same one. And it has difficulties in collecting actual underwater data. In this paper, we synthesized active target echoes based on ray tracing algorithm using target model having 3-dimensional highlight distribution. Then, Fractional Fourier transform was applied to synthesized target echoes to extract feature vector. Recognition experiment was performed using probabilistic neural network classifier.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.10
• /
• pp.2461-2469
• /
• 1997

Target detection and recognition problems, in which neural networks are widely used, require translation invariant and real-time processing in addition to the requirements that general pattern recognition problems need. This paper presents a novel architecture that meets the requirements and explains effective methodology to train the network. The proposed neural network is an architectural extension of the shared-weight neural network that is composed of the feature extraction stage followed by the pattern recognition stage. Its feature extraction stage performs correlational operation on the input with a weight kernel, and the entire neural network can be considered a nonlinear correlation filter. Therefore, the output of the proposed neural network is correlational plane with peak values at the location of the target. The architecture of this neural network is suitable for implementing with parallel or distributed computers, and this fact allows the application to the problems which require realtime processing. Net training methodology to overcome the problem caused by unbalance of the number of targets and non-targets is also introduced. To verify the performance, the proposed network is applied to detection and recognition problem of a specific automobile driving around in a parking lot. The results show no false alarms and fast processing enough to track a target that moves as fast as about 190 km per hour.

• Journal of Satellite, Information and Communications
• /
• v.12 no.1
• /
• pp.64-71
• /
• 2017

In this paper, Kalman filter and RRWA algorithm are used to estimate the accurate target in IR-UWB (Impulse-Radio Ultra Wideband) radar system, which enables accurate location recognition of indoors and outdoors with low cost and low power consumption. In the signal reflected by the target, unnecessary signals exist in addition to the target signal. We have tried to remove unnecessary signals and to derive accurate target signals and improve performance. The location of the targets is estimated in real time with one transmitting antenna and one receiving antenna. The Kalman filter was used to remove the background noise and the RRWA algorithm was used to remove the reflected signal. In this paper, we think that it will be useful to study the accurate distance estimation and tracking in future target estimation.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.23 no.8
• /
• pp.1002-1009
• /
• 2012

This paper shows the research about radar target recognition using the measured radar signals from MSU(Michgan State University) bistatic radar system. In this research, we first did the bistatic measurements at $30^{\circ}$, $60^{\circ}$, $90^{\circ}$ using F-14, Mig-29, and F-22 scale models. Then, we extract the target feature vectors using time-frequency analysis methods such as STFT(Short Time Fourier Transform) and CWT(Continous Wavelet Transform) and perform the target classification test using MLP(Multi-layerd Perceptron) neural network. The results show that the target classification performance is too much dependent on the bistatic angles and the best performance is obtained at the $60^{\circ}$ bistatic angle.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.36 no.6
• /
• pp.525-534
• /
• 2018

Camera calibration is the process of determining the parameters such as the focal length of a camera, the position of a principal point, and lens distortions. For this purpose, images of checkerboard have been mainly used. When targets were automatically recognized in checkerboard image, the existing studies had limitations in that the user should have a good understanding of the input parameters for recognizing the target or that all checkerboard should appear in the image. In this study, a methodology for automatic target recognition was proposed. In this method, even if only a part of the checkerboard image was captured using rectangles including eight blobs, four each at the central portion and the outer portion of the checkerboard, the index of the target can be automatically assigned. In addition, there is no need for input parameters. In this study, three conditions were used to automatically extract the center point of the checkerboard target: the distortion of black and white pattern, the frequency of edge change, and the ratio of black and white pixels. Also, the direction and numbering of the checkerboard targets were made with blobs. Through experiments on two types of checkerboards, it was possible to automatically recognize checkerboard targets within a minute for 36 images.

• Journal of Institute of Control, Robotics and Systems
• /
• v.14 no.12
• /
• pp.1205-1211
• /
• 2008

This paper describes methods to estimate 3-D position of target with respect to reference frame through monocular image from unmanned aerial vehicle (UAV). 3-D position of target is used as information for surveillance, recognition and attack. In this paper. 3-D position of target is estimated to make guidance and control law, which can follow target, user interested. It is necessary that position of target is measured in image to solve 3-D position of target. In this paper, kalman filter is used to track and output position of target in image. Estimation of target's 3-D position is possible using result of image tracking and information of UAV and camera. To estimate this, two algorithms are used. One is methode from arithmetic derivation of dynamics between UAV, carmer, and target. The other is LPV (Linear Parametric Varying). These methods have been run on simulation, and compared in this paper.

• Journal of Korea Multimedia Society
• /
• v.18 no.4
• /
• pp.557-564
• /
• 2015

There are many prior studies for more natural Human-Computer Interaction. Until now, the efforts is continued in order to recognize motions in various directions. In this paper, we suggest a user-specific recognition by hit detection method using Kinect camera and human proportion. This algorithm extracts the user-specific valid recognition rage after recognizing the user's body initially. And it corrects the difference in horizontal position between the user and Kinect, so that we can estimate a action of user by matching cursor to target using only one frame. Ensure that efficient hand recognition in the game to take advantage of this method of suggestion.

• The Journal of Korea Robotics Society
• /
• v.12 no.4
• /
• pp.432-440
• /
• 2017

In this paper, we propose a jellyfish distribution recognition and monitoring system using a UAV (unmanned aerial vehicle). The UAV was designed to satisfy the requirements for flight in ocean environment. The target jellyfish, Aurelia aurita, is recognized through convolutional neural network and its distribution is calculated. The modified deep neural network architecture has been developed to have reliable recognition accuracy and fast operation speed. Recognition speed is about 400 times faster than GoogLeNet by using a lightweight network architecture. We also introduce the method for selecting candidates to be used as inputs to the proposed network. The recognition accuracy of the jellyfish is improved by removing the probability value of the meaningless class among the probability vectors of the evaluated input image and re-evaluating it by normalization. The jellyfish distribution is calculated based on the unit jellyfish image recognized. The distribution level is defined by using the novelty concept of the distribution map buffer.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.11
• /
• pp.1489-1495
• /
• 1995

In this paper, we propose recurrent neural prediction models (RNPM), recurrent neural networks trained as a nonlinear predictor of speech, as a new connectionist model for speech recognition. RNPM modulates its mapping effectively by internal representation, and it requires no time alignment algorithm. Therefore, computational load at the recognition stage is reduced substantially compared with the well known predictive neural networks (PNN), and the size of the required memory is much smaller. And, RNPM does not suffer from the problem of deciding the time varying target function. In the speaker dependent and independent speech recognition experiments under the various conditions, the proposed model was comparable in recognition performance to the PNN, while retaining the above merits that PNN doesn't have.