|
Chinese character detection using modified Single Shot Multibox Detector |
|
Junhwan Ryu, Sungho Kim(Yeungnam University, Korea) |
| This paper deals with the research for detecting Chinese characters. It is based on SSD, which is a conventional single shot detector, and uses data augmentation and network fine tuning to detect Chinese characters. |
|
|
Automatic Sketch Colorization using DCGAN |
|
Hwan Heo, Youngbae Hwang(Korea Electronics Technology Institute, Korea) |
| In general, the manual coloring task from the black-white sketch is complicated and time-consuming. Furthermore, in the case of coloring which is a repetition of a similar pattern, it can be seen that manual coloring is less efficient. Therefore, the technique of automatically coloring from black-white sketch can become a practical application. We propose automatic sketch colorization by using U-Net and deep convolutional generative adversarial network (DCGAN) in the generative model. Experimental results on test set show various results including errors depend on test images. |
|
|
3D Human Pose Estimation Network Using Voxel Data |
|
Han-Mu Park, Ju Hong Yoon(Korea Electronics Technology Institute, Korea), Youngbae Hwang(KETI, Korea) |
| In this paper, we introduce an efficient 3D human pose estimation method using the deep learning technology. The proposed network has the slimmer architecture than the state-of-the-art network, and shows accurate 3D human pose estimation in the voxel space. |
|
|
Cost-function evaluation for intensity-based 2D-3D registration of broken femur bone |
|
Asaduz Zaman, Seong Young Ko(Chonnam National University, Korea) |
| 2D-3D registration is the process to find the relationship between pre-op 3D data and intra-op 2D images. For exact and robust convergence, it is important to find an appropriate cost-function. In this study, three different cost-functions, normalized cross correlation (NCC), normalized mutual information, and gradient difference, have been scrutinized based on their convergence trend, execution time, and effective capture range to find the best one to be used for experiments with phantom X-ray image with CT data. The NCC is found to be the best cost function among the studied cost-functions. |
|
|
A License Plate Recognition using Neural Network and Autonomous Mobile Robot in Intelligent Parking Lot Management System |
|
Wangheon Lee(Hansei University, Korea) |
| We developed not only the ROS based autonomous mobile robot (AMR) in order to control the parking lot management (PLM), but also the central control system (CCS) of the PLM through heterogenous wireless network. The images are extracted from information of RGB-D taken by the PLM for the license plate recognition (LPR) and the LPR is based on the support vector machine while maneuvering the autonomous management of the parking lot. Through applying the developed autonomous mobile robot PLM and the SVM based LPR to the real intelligent parking lot and analyses of the extracted data. |
| |