We propose a method for unsupervised video object segmentation by transferring the knowledge encapsulated in image-based instance embedding networks. The instance embedding network produces an embedding vector for each pixel that enables identifying all pixels belonging to the same object. Though trained on static images, the instance embeddings are stable over consecutive video frames, which allows us to link objects together over time. Thus, we adapt the instance networks trained on static images to video object segmentation and incorporate the embeddings with objectness and optical flow features, without model retraining or online fine-tuning. The proposed method outperforms state-of-the-art unsupervised segmentation methods in the DAVIS dataset and the FBMS dataset.
16 May 2017 in IEEE Journal of Oceanic Engineering
Having an accurate mathematical model is essential in design, control, and navigation process of an autonomous underwater vehicle (AUV). Due to the modeling simplifications, the available mathematical models suffer from the uncertainty of their parameters and they usually need an identification phase for improving the modeling accuracy. In case of AUVs, the hydrodynamic coefficients of the model play an important role and they should be identified using the available experimental data. In this paper, a 6-DoF dynamic model of an AUV is presented and then some of its parameters including viscous damping, the body lift, and control input coefficients that have the highest effects on the modeling error are identified by augmented state model method. The extended Kalman filter (EKF), cubature Kalman filter (CKF), and transformed unscented Kalman filter (TUKF) are used as the estimation filters. To verify and compare the three estimation filters with consideration of the predetermined hydrodynamic coefficients and spiral maneuver AUV results, these three methods are evaluated. The results indicate that the TUKF identifies the best hydrodynamic model due to solving both the CKF nonlocal sampling problem and the EKF linearization problem.
4 December 2017 in IEEE Journal of Oceanic Engineering
In navigation and motion control of an autonomous vehicle, estimation of attitude and heading is an important issue, especially when the localization sensors such as global positioning system (GPS) are not available and the vehicle is navigated by the dead reckoning (DR) strategies. In this paper, based on a new modeling framework, an extended Kalman filter (EKF) is utilized for estimation of attitude, heading, and gyroscope sensor bias using a low-cost microelectromechanical system (MEMS) inertial sensor. The algorithm is developed for accurate estimation of attitude and heading in the presence of external disturbances including external body accelerations and magnetic disturbances. In this study, using the proposed attitude and heading reference system (AHRS) and an identified surge dynamic model of an autonomous underwater vehicle, a low-cost model-aided DR navigation system has been designed for an AUV. The proposed algorithm application is evaluated by experimental tests in different acceleration bound and existence of external magnetic disturbances for an AUV. The results indicate that the roll, pitch, and heading are estimated by mean value errors about 0.35°, 0.29°, and 1.27°, respectively. Moreover, they indicate that a relative navigation error about 8% of the traveling distance can be achieved using the developed approach in during GPS outages.
11 December 2012 in Journal of Intelligent Manufacturing
One of the most important problems of total hip replacement is aseptic loosening of the femoral component, which is related to the changes of the stress distribution pattern after implantation of the prosthesis. Stress shielding of the femur is recognized as a primary factor in aseptic loosening of hip replacements. Utilizing different materials is one of the ordinary solutions for that problem, but using functionally graded materials (FGMs) could be better than the conventional solutions. This research work aims at investigating different porous FGM implants and a real femoral bone by a 3D finite element method. The results show that a neutral functionally graded prosthesis cannot extraordinarily make changes in the stress pattern of bone and prosthesis, but an increasing functionally graded prosthesis leads a lower level of stress in the prosthesis, and a decreasing functionally graded prosthesis can properly reduce the stress shielding among these three architectures. Due to the absence of similar results in the specialized literature, this paper is likely to fill a gap in the state-of-the-art bio-implants, and provide pertinent results that are instrumental in the design of porous femoral prostheses under normal walking loading conditions.
1 January 2022 in Composites Science and Technology