Pedestrian trajectory prediction in dynamic scenes remains a challenging and critical problem in numerous applications, such as self-driving cars and socially aware robots. Challenges concentrate on capturing pedestrians' motion patterns and social interactions, as well as handling the future uncertainties. Recent studies focus on modeling pedestrians' motion patterns with recurrent neural networks, capturing social interactions with pooling- or graph-based methods, and handling future uncertainties by using the random Gaussian noise as the latent variable. However, they do not integrate specific obstacle avoidance experiences (OAEs) that may improve prediction performance. For example, pedestrians' future trajectories are always influenced by others in front. Here, we propose the Graph-based Trajectory Predictor with Pseudo-Oracle (GTPPO), an encoder-decoder-based method conditioned on pedestrians' future behaviors. Pedestrians' motion patterns are encoded with a long short-term memory unit, which introduces temporal attention to highlight specific time steps. Their interactions are captured by a graph-based attention mechanism, which draws OAE into the data-driven learning process of graph attention. Future uncertainties are handled by generating multimodal outputs with an informative latent variable. Such a variable is generated by a novel pseudo-oracle predictor, which minimizes the knowledge gap between historical and ground-truth trajectories. Finally, the GTPPO is evaluated on ETH, UCY, and Stanford Drone datasets, and the results demonstrate state-of-the-art performance. Besides, the qualitative evaluations show successful cases of handling sudden motion changes in the future. Such findings indicate that GTPPO can peek into the future.

Forecasting pedestrian trajectories in dynamic scenes remains a critical problem in various applications, such as autonomous driving and socially aware robots. Such forecasting is challenging due to human-human and human-object interactions and future uncertainties caused by human randomness. Generative model-based methods handle future uncertainties by sampling a latent variable. However, few studies explored the generation of the latent variable. In this work, we propose the trajectory predictor with pseudo Oracle (TPPO), which is a generative model-based trajectory predictor. The first pseudo oracle is pedestrians’ moving directions, and the second one is the latent variable estimated from ground truth trajectories. A social attention module is used to aggregate neighbors’ interactions based on the correlation between pedestrians’ moving directions and future trajectories. This correlation is inspired by the fact that pedestrians’ future trajectories are often influenced by pedestrians in front. A latent variable predictor is proposed to estimate latent variable distributions from observed and ground-truth trajectories. Moreover, the gap between these two distributions is minimized during training. Therefore, the latent variable predictor can estimate the latent variable from observed trajectories to approximate that estimated from ground-truth trajectories. We compare the performance of TPPO with related methods on several public datasets. Results demonstrate that TPPO outperforms state-of-the-art methods with low average and final displacement errors. The ablation study shows that the prediction performance will not dramatically decrease as sampling times decline during tests.

The number of sources present in a mixture is crucial information often assumed to be known or detected by source counting. The exiting methods for source counting in underdetermined blind speech separation (UBSS) suffer from the overlapping between sources with low W-disjoint orthogonality (WDO). To address this issue, we propose to fit the direction of arrival (DOA) histogram with multiple von-Mises density (VM) functions directly and form a sparse recovery problem, where all the source clusters and the sidelobes in the DOA histogram are fitted with VM functions of different spatial parameters. We also developed a formula to perform the source counting taking advantage of the values of the sparse source vector to reduce the influence of sidelobes. Experiments are carried out to evaluate the proposed source counting method and the results show that the proposed method outperforms two well-known baseline methods.

Deep learning methods based on neural network models have been widely applied to bearing fault classification. Although they can achieve high accuracy, they also come with significant complexity. Bearing faults often generate impact vibrations, which produce regular fault characteristic peaks on the envelope spectrum. This paper utilizes the differences in frequency and intensity of the envelope spectrum characteristic peaks under different bearing fault conditions as fault features. By combining these features with the simple and efficient Naive Bayes classifier for fault diagnosis, the algorithm complexity is reduced from the perspective of feature extraction and fault identification. The proposed method was validated using bearing fault data from the Case Western Reserve University (CWRU) dataset and the Machinery Fault Prediction Technology (MFPT) dataset. The results show that the method can classify bearing faults and achieve accurate diagnostic results. The average diagnostic accuracy for the four groups from these two datasets was 99.90% and 99.65%, respectively. The Naive Bayes classification algorithm was compared with classic algorithms in terms of classification accuracy and classification time. Additionally, the algorithm was compared with recent bearing fault diagnosis methods using the CWRU dataset in terms of algorithm complexity. The complexity of the proposed algorithm is only O(N(2280)), which is lower than that of other bearing fault diagnosis methods, where N represents the number of fault samples. This demonstrates that the method significantly reduces algorithm complexity while ensuring accuracy, improving diagnostic efficiency, enhancing the timeliness of real-time industrial bearing fault diagnosis, and reducing hardware setup and operating costs.

Nowadays, Magnetic and Inertial Measurement Units (MIMUs) have become useful tools for determining the attitude of moving objects. An Attitude Calculation Algorithm (ACA) is programmed into a MIMU, which determines the attitude by fusing three built-in sensors, namely, gyroscope, accelerometer, and magnetometer. Attitude is often represented by Euler angles, quaternion, or rotation matrix. Despite that great development has been achieved, how to eliminate the negative effect of magnetic disturbance on the pitch and roll estimation is still difficult for a ACA. This paper aims to propose an Extended Kalman Filter (EKF) based approach for problem solving. Inspired by the fact that the Euler angles are computed by using some elements of the rotation matrix, a Diagonalizable Regulation Matrix (DRM) is multiplied to the right half part of the gain matrix (which is multiplied to the residual of the magnetometer measurement in the update stage of EKF). The determination of DRM is based on the principle that these elements can be appropriately regulated by the DRM so as to make the magnetometer measurement only participate in the computation of azimuth. No further revision to EKF is made except the Kalman gain matrix, hence the proposed EKF behaves similar as traditional EKF, except its ability of eliminating the negative effect of magnetic disturbance on the pitch and roll estimation. The superiority of the proposed EKF is validated by comparing it with four other competing algorithms through simulation and experimental experiments.

Magnetic and Inertial Measurement Units (MIMUs) are mainly used for determining the attitude of moving bodies. The Extended Kalman Filter (EKF) and the Complementary Filters (CFs) are the most commonly applied algorithms for calculating the attitude of the body that a MIMU is attached. Despite its much higher accuracy than CFs, the time-consuming EKF may not be competent for the work of attitude calculation if a microcomputer with limited computation power is used. To any attitude calculating algorithm, its Posterior Attitude (PA) is computed as: PA = prior attitude + Gain Matrix (GM) × innovation. The prior attitude and the innovation are derived from the measurements of the gyroscope, the accelerometer and the magnetometer. Detailed analysis in this study found that the reason for the performance difference between EKF and CFs is their different structures of the GMs, which promotes the authors to construct a new GM. Based on some reasonable assumptions, the GM of EKF is simplified under the principle that the computation accuracy of PA is decreased as little as possible, but its computation burden is reduced as much as possible. After the replacement of the GM of the EKF by the innovatively designed GM, a new complementary filter, named TCF, is proposed. The simulation and experimental results show that TCF performs better than two CFs and is close to EKF in terms of attitude estimation accuracy, but uses nearly the same computation time as the two CFs, and about half of the computation time required by EKF.