Depth image enhancement techniques can help to improve image quality and facilitate computer vision tasks. Traditional image-enhancement methods, which are typically based on integer-order calculus, cannot exploit the textural information of an image, and their enhancement effect is limited. To solve this problem, fractional differentiation has been introduced as an innovative image-processing tool. It enables the flexible use of local and non-local information by taking into account the continuous changes between orders, thereby improving the enhancement effect. In this study, a fractional differential is applied in depth image enhancement and used to establish a novel algorithm, named the fractional differential-inverse-distance-weighted depth image enhancement method. Experiments are performed to verify the effectiveness and universality of the algorithm, revealing that it can effectively solve edge and hole interference and significantly enhance textural details. The effects of the order of fractional differentiation and number of iterations on the enhancement performance are examined, and the optimal parameters are obtained. The process data of depth image enhancement associated with the optimal number of iterations and fractional order are expected to facilitate depth image enhancement in actual scenarios.

Local feature descriptors are a critical problem in computer vision; the majority of current approaches find it difficult to achieve a balance between descriptiveness, robustness, compactness, and efficiency. This paper proposes the local discrete feature histogram (LDFH), a novel local feature descriptor, as a solution to this problem. The LDFH descriptor is constructed based on a robust local reference frame (LRF). It partitions the local space based on radial distance and calculates three geometric features, including the normal deviation angle, polar angle, and normal lateral angle, in each subspace. These features are then discretized to generate three feature statistical histograms, which are combined using a weighted fusion strategy to generate the final LDFH descriptor. Experiments on public datasets demonstrate that, compared with the existing methods, LDFH strikes an excellent balance between descriptiveness, robustness, compactness, and efficiency, making it suitable for various scenes and sensor datasets.

The target is often obstructed by obstacles with the dynamic tracking scene, leading to a loss of target information and a decrease in tracking accuracy or even complete failure. To address these challenges, we leverage the capabilities of Geiger-mode Avalanche Photodiode (GM-APD) LiDAR to acquire both intensity images and point cloud data for researching a target tracking method that combines the fusion of intensity images and point cloud data. Building upon Kernelized correlation filtering (KCF), we introduce Fourier descriptors based on intensity images to enhance the representational capacity of target features, thereby achieving precise target tracking using intensity images. Additionally, an adaptive factor is designed based on peak sidelobe ratio and intrinsic shape signature to accurately detect occlusions. Finally, by fusing the tracking results from Kalman filter and KCF with adaptive factors following occlusion detection, we obtain location information for the central point of the target. The proposed method is validated through simulations using the KITTI tracking dataset, yielding an average position error of 0.1182m for the central point of the target. Moreover, our approach achieves an average tracking accuracy that is 21.67% higher than that obtained by Kalman filtering algorithm and 7.94% higher than extended Kalman filtering algorithm on average.

Trajectory prediction is critical for safe robot navigation, yet standard deep learning models predominantly rely on the Mean Squared Error (MSE) criterion. While effective under ideal conditions, MSE-based optimization is inherently fragile to non-Gaussian impulsive noise—such as sensor glitches and occlusions—common in real-world deployment. To address this limitation, this paper proposes MEE-LSTM, a robust forecasting framework that integrates Long Short-Term Memory networks with the Minimum Error Entropy (MEE) criterion. By minimizing Renyi’s quadratic entropy of the prediction error, our loss function introduces an intrinsic “gradient clipping” mechanism that effectively suppresses the influence of outliers. Furthermore, to overcome the convergence challenges of fixed-kernel information theoretic learning, we introduce a Silverman-based Adaptive Annealing (SAA) strategy that dynamically regulates the kernel bandwidth. Extensive evaluations on the ETH and UCY datasets demonstrate that MEE-LSTM maintains competitive accuracy on clean benchmarks while exhibiting superior resilience in degraded sensing environments. Notably, we identify a “Scissor Plot” phenomenon under stress testing: in the presence of 20% impulsive noise, the proposed model maintains a stable Average Displacement Error (ADE “≈” 0.51 m), whereas MSE baselines suffer catastrophic degradation (ADE > 2.1 m), representing a 75.7% improvement in robustness. This work provides a statistically grounded paradigm for reliable causal inference in hostile robotic perception.

To address the slow response, low pointing accuracy, and degraded stability caused by the coexistence of fractional-order dynamics, dual-axis coupling, and system delay in liquid crystal optical phased array (LCOPA) beam control, a fractional-order composite control method is proposed. First, the LCOPA is modeled as a two-input two-output system, and a fractional-order time-delay dynamic model is established. A composite control architecture integrating a fractional-order controller, a diagonal decoupling compensator, and a Smith predictor is then designed to improve dynamic response, suppress cross-axis coupling, and mitigate delay-induced performance degradation. A multi-strategy improved sparrow search algorithm is used to optimize the dual-channel fractional-order controller parameters, and closed-loop stability is verified using characteristic roots and the Matignon criterion. Simulation results show that the settling times of both axes are 11.50 ms, reduced by 62.54%/62.30%, 57.09%/56.11%, and 51.68%/48.89% compared with the results for PID, FOPID, and ADRC, respectively. The maximum cross-axis deviations are reduced to 0.0036° and 0.0030°, respsectively. Experiments further show that the mean absolute pointing errors of the X- and Y-axes decrease from 0.0466° and 0.0362° to 0.0057° and 0.0045°, while the RMSEs under external disturbances are 0.0060° and 0.0044°, respectively. These results demonstrate improved response speed, pointing accuracy, decoupling performance, and disturbance rejection.