Abstract Fisheye cameras are widely used in surveillance, automotive systems, virtual reality, and panoramic photography due to their wide‐angle perspective. However, images captured by fisheye cameras suffer from significant geometric distortions, affecting image analysis and understanding. This distortion bends straight lines into curves, resulting in a barrel‐shaped appearance of the image. To mitigate these effects and transform fisheye images into a regular perspective, fisheye image correction is necessary, enabling more accurate and reliable performance in applications like object recognition, 3D reconstruction, and visual navigation. While convolutional neural networks based fisheye image correction has progressed, it has not fully utilized the global distribution and local symmetry of distortions due to the limitations of fixed receptive fields. This paper introduces a new model based on the Swin Transformer that effectively utilizes both global and local distortion features to adapt automatically to fisheye lens distortions. It also incorporates image restoration functionality to enhance texture details in the corrected images. A novel approach to synthetic dataset generation is proposed to improve the network's generalization capabilities.

We have been developing a paradigm that we call learning-from-observation for a robot to automatically acquire a robot program to conduct a series of operations, or for a robot to understand what to do, through observing humans performing the same operations. Since a simple mimicking method to repeat exact joint angles or exact end-effector trajectories does not work well because of the kinematic and dynamic differences between a human and a robot, the proposed method employs intermediate symbolic representations, tasks, for conceptually representing what-to-do through observation. These tasks are subsequently mapped to appropriate robot operations depending on the robot hardware. In the present work, task models for upper-body operations of humanoid robots are presented, which are designed on the basis of Labanotation. Given a series of human operations, we first analyze the upper-body motions and extract certain fixed poses from key frames. These key poses are translated into tasks represented by Labanotation symbols. Then, a robot performs the operations corresponding to those task models. Because tasks based on Labanotation are independent of robot hardware, different robots can share the same observation module, and only different task-mapping modules specific to robot hardware are required. The system was implemented and demonstrated that three different robots can automatically mimic human upper-body operations with a satisfactory level of resemblance.

The present study used a flat keyboard without moving keys and enabled with haptic keyclick feedback to examine the effect of haptic keyclick feedback on touch typing performance. We investigated, with well-controlled stimuli and a within-participant design, how haptic keyclick feedback might improve typing performance in terms of typing speed, typing efficiency and typing errors. Of the three kinds of haptic feedback we tested, all increased typing speed and decreased typing errors compared to a condition without haptic feedback. We did not find significant differences among the types of haptic feedback. We also found that auditory keyclick feedback alone is not as effective as haptic keyclick feedback, and the addition of auditory feedback to haptic feedback does not lead to any significant improvement in typing performance. We also learned that global haptic keyclick feedback simulated through local keyclick feedback on each key (as opposed to haptic feedback all over the keyboard) might have the additional and unexpected benefit of helping a typist to locate keys on a keyboard. Furthermore, the participants preferred auditory or haptic keyclick feedback to no feedback, and haptic feedback restricted to the typing finger alone is preferred to that over a larger area of the keyboard.