|
Calibration and 3D Reconstruction of Images Obtained Using Spherical Panoramic Camera |
|
Hirokazu Madokoro(Iwate Prefectural University, Japan), Satoshi Yamamoto, Yo Nishimura, Stephanie Nix(Akita Prefectural University, Japan), Hanwool Woo(The University of Tokyo, Japan), Kazuhito Sato(Akita Prefectural University, Japan) |
| This study was conducted to develop a 3D reconstruction procedure for application to crop monitoring. For 3D construction of a similar target object, we compared images obtained from two camera types: a compact digital camera (CDC) and a spherical panoramic camera (SPC). Experimentally obtained results demonstrated that the 3D reconstruction of a target object was improved after calibration compared with that before calibration. Moreover, we conducted an application experiment using a tree in an outdoor environment as a trial of practical use at a farm. |
|
|
Image-Goal Navigation via Metric Mapping and Keypoint based Reinforcement Learning |
|
Jae Seok Heo, Yunho Choi, Songhwai Oh(Seoul National University, Korea) |
| In this paper, the agent is asked to find the viewpoint of a given image-goal in an unseen environment. We propose using a hierarchical policy structure which uses a high-level policy to guide the agent more efficiently to the image-goal. Our method uses metric maps built by RGB images via Active Neural SLAM network and matched keypoints between the observation and the image-goal extracted from place recognition networks. We have trained a policy network which uses these information as inputs, and output is a global goal which guides the agent to the image-goal. It is shown that using this method gives improved performance compared to previous related methods. |
|
|
An Integrated System Design Interface for Operating 8-DoF Robotic Arm |
|
Vishal Gattani, Madhav Rao(International Institute of Information Technology Bangalore, India) |
| This study examines the system integration of a human “pilot” with robotics middleware to drive an 8 degree of freedom robotic upper limb to generate human-like motion for telerobotic applications. The developed architecture encompasses a pipeline execution design using Blender Game Engine (BGE) including the acquisition of real human movements via the Microsoft Kinect V2, interfaced with a modeled virtual arm, and replication of similar arm movements on the physical robotic arm. A simple and intuitive kinematic modeling and 3D simulation process is presented, which is validated using the arm. |
|
|
High-speed Embedded Optical Flow Measurement System for Real-time Use |
|
Masaya Ozaki, Teruo Yamaguchi(Kumamoto University, Japan) |
| In the development of visual sensors, the calculation speed and accuracy of the measurement are issues that should be improved. Parallel trade off of the processing was used as a means of speeding up the calculation. In this research, we have developed a system that can measure optical flow with an embedded microprocessor. We have conducted two experiments, one in which the camera was fixed and a moving subject was photographed, and the other in which the camera was moved and photographed. Experimented results show that was possible to reduce the calculation time by about one-third for real-time processing, and it was possible to detect the difference in the speed of the camera itself. |
|
|
Effect of the Covariance Matrix of the Kalman Filter used for Optical Flow Estimation |
|
Kohei Otani, Teruo Yamaguchi(Kumamoto University, Japan) |
| We propose a method of observing the velocity of a target object using the spatio-temporal differentiation method. Estimating and recognizing the position and velocity of the target object by motion image processing using a Kalman filter. Then, the velocity of the target object is observed by the spatio-temporal differentiation method which introduced the compensation method. The error of the observed value can be reduced by determining the compensation amount using the predicted value by the Kalman filter. In this study, we examine how to set the covariance matrix of the Kalman filter. |
|
|
GBNet: Gradient Boosting Network for Monocular Depth Estimation |
|
Daechan Han, Yukyung Choi(Sejong University, Korea) |
| This paper proposes a novel self- and semi-supervised monocular depth estimation method, inspired by the gradient boosting method. We design our proposed network to refine the predicted depth map sequentially and gradually generate a high-quality depth map via multi-stack CNN structures. Our method shows the state-of-the-art results for monocular depth estimation on a DDAD (Dense Depth for Autonomous Driving) dataset. |
|