• The Journal of Korea Robotics Society
• /
• v.4 no.4
• /
• pp.298-304
• /
• 2009

This paper describes an algorithm that improves 3D reconstruction result using a multi-sensor fusion disparity map. We can project LRF (Laser Range Finder) 3D points onto image pixel coordinatesusing extrinsic calibration matrixes of a camera-LRF (${\Phi}$, ${\Delta}$) and a camera calibration matrix (K). The LRF disparity map can be generated by interpolating projected LRF points. In the stereo reconstruction, we can compensate invalid points caused by repeated pattern and textureless region using the LRF disparity map. The result disparity map of compensation process is the multi-sensor fusion disparity map. We can refine the multi-sensor 3D reconstruction based on stereo vision and LRF using the multi-sensor fusion disparity map. The refinement algorithm of multi-sensor based 3D reconstruction is specified in four subsections dealing with virtual LRF stereo image generation, LRF disparity map generation, multi-sensor fusion disparity map generation, and 3D reconstruction process. It has been tested by synchronized stereo image pair and LRF 3D scan data.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.22 no.5
• /
• pp.523-531
• /
• 2002

This study is concentrated on the development of high speed multi-object image processing algorithm in real time. Recently, the use of vision system is rapidly increasing in inspection and robot's position control. To apply the vision system, it is necessary to transform the physical coordinate of object into the image information acquired by CCD camera. Thus, to use the application of the vision system to the inspection and robot's position control in real time, we have to know the position of object in the image plane. Particularly, in case of rigid body using multi-cue to identify its shape, the each position of multi-cue must be calculated in an image plane at the same time. To solve these problems, the image processing algorithm on the position of multi-cue is developed.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.273-282
• /
• 2022

MPEG-VCM(Video Coding for Machine) aims to standardize video codec for machines. VCM provides data sets and anchors, which provide reference data for comparison, for several machine vision tasks including object detection, object segmentation, and object tracking. The evaluation template can be used to compare compression and machine vision task performance between anchor data and various proposed video codecs. However, performance comparison is carried out separately for each machine vision task, and information related to performance evaluation of multiple machine vision tasks on a single bitstream is not provided currently. In this paper, we propose a performance evaluation method of a video codec for AI-based multi-tasks. Based on bits per pixel (BPP), which is the measure of a single bitstream size, and mean average precision(mAP), which is the accuracy measure of each task, we define three criteria for multi-task performance evaluation such as arithmetic average, weighted average, and harmonic average, and to calculate the multi-tasks performance results based on the mAP values. In addition, as the dynamic range of mAP may very different from task to task, performance results for multi-tasks are calculated and evaluated based on the normalized mAP in order to prevent a problem that would be happened because of the dynamic range.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.608-611
• /
• 1997

The soccer robot system consists of multi agents, with highly coordinated operation and movements so as to fulfill specific objectives, even under adverse situation. The coordination of the multi-agents is associated with a lot of supplementary work in advance. The associated issues are the position correction, prevention of communication congestion, local information sensing in addition to the need for imitating the human-like decision making. A control structure for soccer robot is designed and several behaviors and actions for a soccer robot are proposed. Variable zone defense as a basic strategy and several special strategies for fouls are applied to SOTY2 team.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.2
• /
• pp.483-503
• /
• 2014

In this paper, human action recognition using pyramid histograms of oriented gradients and collaborative multi-task learning is proposed. First, we accumulate global activities and construct motion history image (MHI) for both RGB and depth channels respectively to encode the dynamics of one action in different modalities, and then different action descriptors are extracted from depth and RGB MHI to represent global textual and structural characteristics of these actions. Specially, average value in hierarchical block, GIST and pyramid histograms of oriented gradients descriptors are employed to represent human motion. To demonstrate the superiority of the proposed method, we evaluate them by KNN, SVM with linear and RBF kernels, SRC and CRC models on DHA dataset, the well-known dataset for human action recognition. Large scale experimental results show our descriptors are robust, stable and efficient, and outperform the state-of-the-art methods. In addition, we investigate the performance of our descriptors further by combining these descriptors on DHA dataset, and observe that the performances of combined descriptors are much better than just using only sole descriptor. With multimodal features, we also propose a collaborative multi-task learning method for model learning and inference based on transfer learning theory. The main contributions lie in four aspects: 1) the proposed encoding the scheme can filter the stationary part of human body and reduce noise interference; 2) different kind of features and models are assessed, and the neighbor gradients information and pyramid layers are very helpful for representing these actions; 3) The proposed model can fuse the features from different modalities regardless of the sensor types, the ranges of the value, and the dimensions of different features; 4) The latent common knowledge among different modalities can be discovered by transfer learning to boost the performance.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.17 no.1
• /
• pp.453-458
• /
• 2016

One of the most common applications of machine vision is quality inspections in automated production. In this study, a welding inspection system that is controlled by a PC and a PLC equipped with a multi-camera setup was developed. The system was designed to measure the primary dimensions, such as the length and width of the welding areas. The TCP/IP protocols and multi-threading techniques were used for parallel control of the optical components and physical distribution. A coaxial light was used to maintain uniform lighting conditions and enhance the image quality of the weld areas. The core image processing system was established through a combination of various algorithms from the OpenCV library. The proposed vision inspection system was fully validated for an actual weld production line and was shown to satisfy the functional and performance requirements.

• Journal of Auto-vehicle Safety Association
• /
• v.3 no.2
• /
• pp.28-33
• /
• 2011

This paper presents driving path estimation algorithm for adaptive cruise control system and advanced emergency braking system using multi-sensor fusion. Through data collection, yaw rate filtering based road curvature and vision sensor road curvature characteristics are analyzed. Yaw rate filtering based road curvature and vision sensor road curvature are fused into the one curvature by weighting factor which are considering characteristics of each curvature data. The proposed driving path estimation algorithm has been investigated via simulation performed on a vehicle package Carsim and Matlab/Simulink. It has been shown via simulation that the proposed driving path estimation algorithm improves primary target detection rate.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.25.2-25
• /
• 2001

This paper describes a mobile robot intended for being employed in a multi-agent system. We have already proposed a multi-agent system which realizes patient-aid by helping a lying patient take a distant object on the table. In this paper, a mobile robot agent is developed and is included in the system. An effective man-machine communication strategy is proposed by use of a vision agent settled on the ceiling. If a human (assumed to be a patient) wishes to take an object distant on the floor, he points to the object. The vision agent detects the direction of his arm by image processing and guesses which object he intends to take. The vision agent asks him if it is what he wants and, if yes, the mobile robot runs to take and bring it to him. The system is overviewed with the explanation of a mobile robot. Some experimental results are shown with discussion.

• Proceedings of the Optical Society of Korea Conference
• /
• 2007.07a
• /
• pp.365-366
• /
• 2007

Micro jetting drop inspection system is essential to measuring micro drop volume. Measuring pico-liter drop volume is useful for new LCD color filter product process that is based on inkjet printing technology. To upgrade the reliability in drop measurement system, we use the auto focusing & multi drop reiteration & blurring average algorism. First of all we used standard mark for gage R&R in the vision system. Finding the most suitable threshold for multi blurring drop, is the main key of this research. Sensitivity of vision system is a standard in measuring the upgrade system level. So, suitable threshold can upgrade the performance of jetting drop inspection system.

• Journal of Institute of Control, Robotics and Systems
• /
• v.4 no.6
• /
• pp.795-801
• /
• 1998

In this paper, we propose a machine vision algorithm for inspecting inserts which are used for milling and turning operations. Major defects of the inserts are breakage and crack on insert surfaces. Among the defects, breakages on the face of the inserts can be detected through three stages of the algorithm developed in this paper. In the first stage, a multi-layer perceptron is used to recognize the inserts being inspected. Edge detection of the insert image is performed in the second stage. Finally, in the third stage breakages on the insert face are identified using Hough transform. The overall algorithm is tested on real specimens and the results show that the algorithm works fairly well.