Multi-head self-attentions (MSAs) in Transformer are low-pass filters, which will tend to reduce high-frequency signals. Convolutional layers (Convs) in Convolutional Neural Network (CNN) are ...high-pass filters, which will tend to capture high-frequency components of the images. Therefore, CNN and Transformer contain complementary information, and the combination of the two is necessary for satisfactory detection results. In this work, we propose a novel framework PileNet that efficiently combine CNN and Transformer for accurate salient object detection (SOD). Specifically in PileNet, we introduce complementary encoder that extracts multi-level complementary saliency features. Next, we simplify the complementary features by adjusting the number of channels for all features to a fixed value. By introducing the multi-level feature aggregation (MLFA) and multi-level feature refinement (MLFR) units, the low- and high-level features can easily be transmitted to feature blocks at various pyramid levels. Finally, we fuse all the refined saliency features in a Unet-like structure from top to bottom and use multi-point supervision mechanism to produce the final saliency maps. Extensive experimental results over five widely used saliency benchmark datasets clearly demonstrate that our proposed model can accurately locate the entire salient objects with clear object boundaries and outperform sixteen previous state-of-the-art saliency methods in terms of a wide range of metrics.
•A high-and-low pass complementary filter is used to generate encoders.•We design an effective multi-level feature refinement unit.•We design a multi-level feature aggregation unit with shared parameters.•A multi-point supervision mechanism is proposed to generate saliency maps.
•Surveyed sensor fusion algorithms for orientation tracking with MIMUs.•Discussed fundamental algorithms like strap-down integration and vector observation.•Discussed advanced algorithms like ...complementary filter and Kalman filter.•Discussed modifications like gain tuning and imperfect measurement rejection.•Provided lessons learned from survey and future research challenges in the field.
Technological developments over the past two decades have resulted in the development of more accurate and lightweight low-cost magnetic and inertial measurement units (MIMUs). These developments have allowed the extensive application of MIMUs in various fields, specifically tracking the 3D orientation of a rigid body. Despite recent technological improvements, measurements from a tri-axial gyroscope, accelerometer, and/or magnetometer inside the MIMU are characterized by uncertainties. Numerous studies have been conducted to address these uncertainties and develop sensor fusion algorithms (SFAs) to estimate the 3D orientation accurately and robustly. This paper contributes to these efforts by providing a survey of the state-of-the-art SFAs for orientation estimation. We surveyed +250 publications, categorized the SFAs with various structures, identified the modifications proposed to improve their performance, and discussed the strengths and weaknesses of these approaches. We found that, while early SFAs were mostly a vector observation algorithm or an extended Kalman filter, to improve the computational efficiency, more recent works have developed SFAs with a complementary filter or complementary Kalman filter structure. At the same time, to improve the performance of the SFAs, several research teams have proposed various modifications to the basic structure of these filters, such as adaptive gain tuning or imperfect measurement rejection. We also provided an outlook on the lessons learned as well as the main challenges related to SFAs and discussed the practical steps toward developing an effective SFA. We have identified the need for benchmarking studies as the main challenge at the moment. This paper is among the first surveys which provide such breadth of coverage across different SFAs for tracking orientation with MIMUs.
Display omitted
This paper proposes a novel quaternion-based attitude estimator with magnetic, angular rate, and gravity (MARG) sensor arrays. A new structure of a fixed-gain complementary filter is designed fusing ...related sensors. To avoid using iterative algorithms, the accelerometer-based attitude determination is transformed into a linear system. Stable solution to this system is obtained via control theory. With only one matrix multiplication, the solution can be computed. Using the increment of the solution, we design a complementary filter that fuses gyroscope and accelerometer together. The proposed filter is fast, since it is free of iteration. We name the proposed filter the fast complementary filter (FCF). To decrease significant effects of unknown magnetic distortion imposing on the magnetometer, a stepwise filtering architecture is designed. The magnetic output is fused with the estimated gravity from gyroscope and accelerometer using a second complementary filter when there is no significant magnetic distortion. Several experiments are carried out on real hardware to show the performance and some comparisons. Results show that the proposed FCF can reach the accuracy of Kalman filter. It successfully finds a balance between estimation accuracy and time consumption. Compared with iterative methods, the proposed FCF has much less convergence speed. Besides, it is shown that the magnetic distortion would not affect the estimated Euler angles.
Focusing on generalized sensor combinations, this paper deals with the attitude estimation problem using a linear complementary filter (CF). The quaternion observation model is obtained via a ...gradient descent algorithm. An additive measurement model is then established according to derived results. The filter is named as the generalized CF where the observation model is simplified as a linear one that is quite different from previous-reported brute-force nonlinear results. Moreover, we prove that representative derivative-based optimization algorithms are essentially equivalent to each other. Derivations are given to establish the state model based on the quaternion kinematic equation. The proposed algorithm is validated under several experimental conditions involving the free-living environment, harsh external field disturbances, and aerial flight test aided by robotic vision. Using the specially designed experimental devices, data acquisition and algorithm computations are performed to give comparisons on accuracy, robustness, time-consumption, and so on with representative methods. The results show that not only the proposed filter can give fast, accurate, and stable estimates in terms of various sensor combinations but also produces robust attitude estimation in the scenario of harsh situations, e.g., irregular magnetic distortion. Note to Practitioners -Multisensor attitude estimation is a crucial technique in robotic devices. Many existing methods focus on the orientation fusion of specific sensor combinations. In this paper, we make the problem more concise. The results given in this paper are very general and can significantly decrease the space consumption and computation burden without losing the original estimation accuracy. Such performance will be of benefit to robotic platforms requiring flexible and easy-to-tune attitude estimation in the future.
Attitude estimation plays a major role in the autonomy of unmanned aerial vehicles and requires fusion of different sensor measurements. This paper describes an adaptive estimation scheme in which ...the weight parameter for the complementary filter (CF) is varied over time. The adaptive mechanism proposed here is inspired from the multiple model adaptive estimation (MMAE) scheme used for varying noise parameters in the Kalman filter structure. In this paper, the linear complementary filters are used as elementary blocks in the MMAE structure and their weights are modified probabilistically to obtain an accurate orientation estimate. It avoids the problem of manual selection of weight factor for complementary filter and provides a robust orientation estimate against varying system dynamics. The proposed MMAE based adaptive CF scheme is modular in nature and is dependent on the residual error between estimated and the measured orientation angle. It is applied on the real world datasets logged from inertial sensors and the performance of MMAE based CF structure is found to work promisingly as compared to the non-linear complementary filter versions and the extended Kalman filter framework.
Attitude estimation is the process of computing the orientation angles of an object with respect to a fixed frame of reference. Gyroscope, accelerometer, and magnetometer are some of the fundamental ...sensors used in attitude estimation. The orientation angles computed from these sensors are combined using the sensor fusion methodologies to obtain accurate estimates. The complementary filter is one of the widely adopted techniques whose performance is highly dependent on the appropriate selection of its gain parameters. This paper presents a novel cascaded architecture of the complementary filter that employs a nonlinear and linear version of the complementary filter within one framework. The nonlinear version is used to correct the gyroscope bias, while the linear version estimates the attitude angle. The significant advantage of the proposed architecture is its independence of the filter parameters, thereby avoiding tuning the filter's gain parameters. The proposed architecture does not require any mathematical modeling of the system and is computationally inexpensive. The proposed methodology is applied to the real-world datasets, and the estimation results were found to be promising compared to the other state-of-the-art algorithms.
With the advancements in micro-electromechanical systems (MEMS) technologies, magnetic and inertial sensors are becoming more and more accurate, lightweight, smaller in size as well as low-cost, ...which in turn boosts their applications in human movement analysis. However, challenges still exist in the field of sensor orientation estimation, where magnetic disturbance represents one of the obstacles limiting their practical application. The objective of this paper is to systematically analyze exactly how magnetic disturbances affects the attitude and heading estimation for a magnetic and inertial sensor. First, we reviewed four major components dealing with magnetic disturbance, namely decoupling attitude estimation from magnetic reading, gyro bias estimation, adaptive strategies of compensating magnetic disturbance and sensor fusion algorithms. We review and analyze the features of existing methods of each component. Second, to understand each component in magnetic disturbance rejection, four representative sensor fusion methods were implemented, including gradient descent algorithms, improved explicit complementary filter, dual-linear Kalman filter and extended Kalman filter. Finally, a new standardized testing procedure has been developed to objectively assess the performance of each method against magnetic disturbance. Based upon the testing results, the strength and weakness of the existing sensor fusion methods were easily examined, and suggestions were presented for selecting a proper sensor fusion algorithm or developing new sensor fusion method.
•Compared 36 sensor fusion algorithms for orientation tracking with MIMUs.•Implemented optimal adaptive gain tuning for sensor fusion algorithms.•Shared sample data and all codes of the implemented ...sensor fusion algorithms.•Identified sensor fusion algorithms with the highest accuracy in various families.
Lightweight and low-cost wearable magnetic and inertial measurement units (MIMUs) have found numerous applications, such as aerial vehicle navigation or human motion analysis, where the 3D orientation tracking of a rigid body is of interest. However, due to the errors in measurements of gyroscope, accelerometer, and/or magnetometer inside a MIMU, numerous studies have proposed sensor fusion algorithms (SFAs) to estimate the 3D orientation accurately and robustly. This paper contributes to these efforts by performing an experimental comparison among a variety of SFAs. Notably, we compared the estimated orientation of 36 SFAs from the complementary filter and linear/extended/complementary/unscented/cubature Kalman filter families with the reference orientation obtained from a camera motion-capture system. The experimental study included data collection with a foot-worn MIMU where nine participants performed various short- and long-duration tasks. We shared the codes and sample of data in https://www.ncbl.ualberta.ca/codes to enable other researchers to compare their works with the literature toward creating a comprehensive online repository for SFAs. To perform a fair comparison, we used the Particle Swarm Optimization routine to find the optimal adaptive gain tuning scheme for each SFAs, as recommended in the literature. Our experimental results showed that gyroscope static bias removal, in general, showed to be effective in reducing the estimation error of SFAs, specifically during long-duration trials. Moreover, our experimental results identified the SFAs with the highest accuracy from each family. We also reported the execution times for the selected SFAs from each family. This paper is among the first experimental comparison studies which provide such breadth of coverage across various SFAs for tracking orientation with MIMUs.
Attitude estimation is a crucial aspect for navigation and motion control of autonomous vehicles. This concept is particularly true in the case of unavailability of localization sensors when ...navigation and control rely on dead reckoning strategies; in this case, indeed, the orientation estimate is also used along with speed measurements to update the position estimate. Among the different approaches proposed in the literature, the de facto state of the art in this field is represented by nonlinear complementary filters: they fuse the measurements of angular rate obtained through gyroscopes, and a measurement of gravity and Earth's magnetic field vectors respectively obtained through accelerometers and magnetometers. This paper is focused on an attitude estimation strategy for autonomous underwater vehicles (AUV). The proposed novelty includes the identification of some critical issues that arise when AUV attitude estimation algorithms are applied in practice. They are mainly due to the use of low-accuracy low-cost microelectromechanical systems (MEMS) sensors and on different sources of magnetic disturbances. Some strategies to overcome the identified issues are proposed, including the integration of a single-axis fiber optic gyroscope (FOG) that ensures a considerable performance improvement with a moderate cost increase. The proposed strategies for detection of issues and sensor fusion have been experimentally tested and validated in a real application scenario estimating the attitude of an AUV performing a lawn mower path. The expected performance improvement is confirmed; the obtained results are described and analyzed in this paper.
This paper proposes a foot-mounted zero velocity update (ZVU) aided inertial measurement unit (IMU) filtering algorithm for pedestrian tracking in indoor environment. The algorithm outputs are the ...foot kinematic parameters that include foot orientation, position, velocity, acceleration, and gait phase. The foot motion filtering algorithm incorporates methods for orientation estimation, gait detection, and position estimation. A novel complementary filter is introduced to better preprocess the sensor data from a foot-mounted IMU containing triaxial angular rate sensors, accelerometers, and magnetometers and to estimate the foot orientation without resorting to global positioning system data. A gait detection is accomplished using a simple states detector that transitions between states based on acceleration and angular rate measurements. Once foot orientation is computed, position estimates are obtained using integrating acceleration and velocity data, which has been corrected at step stance phase for drift using an implemented ZVU algorithm, leading to a position accuracy improvement. We show our findings experimentally by using of a commercial IMU during regular human walking trials in a typical public building. Experiment results show that the positioning approach achieves approximately a position accuracy around 0.4% and improves the performance regarding recent works of literature.