Real-time Performance Research Articles

LS-VIT: Vision Transformer for action recognition based on long and short-term temporal difference

Over the past few years, a growing number of researchers have dedicated their efforts to focusing on temporal modeling. The advent of transformer-based methods has notably advanced the field of 2D image-based vision tasks. However, with respect to 3D video tasks such as action recognition, applying temporal transformations directly to video data significantly increases both computational and memory demands. This surge in resource consumption is due to the multiplication of data patches and the added complexity of self-aware computations. Accordingly, building efficient and precise 3D self-attentive models for video content represents as a major challenge for transformers. In our research, we introduce an Long and Short-term Temporal Difference Vision Transformer (LS-VIT). This method incorporates short-term motion details into images by weighting the difference across several consecutive frames, thereby equipping the original image with the ability to model short-term motions. Concurrently, we integrate a module designed to understand long-term motion details. This module enhances the model's capacity for long-term motion modeling by directly integrating temporal differences from various segments via motion excitation. Our thorough analysis confirms that the LS-VIT achieves high recognition accuracy across multiple benchmarks (e.g., UCF101, HMDB51, Kinetics-400). These research results indicate that LS-VIT has the potential for further optimization, which can improve real-time performance and action prediction capabilities.

Traffic flow parameter estimation for complex environments based on improved YOLOv8

Estimation of traffic flow parameters based on computer vision is still a very popular challenge. In particular, the detection problems such as occlusion, small targets, luminance variations, and lighting jitter, etc., are caused by dense traffic scenes and complex weather conditions. Previous studies about detecting and tracking methods focused on daytime and nighttime or single weather condition, but the performance of this model will decline dramatically due to complex conditions. In this paper, a framework for estimating traffic flow parameters in complex environments is presented, in which Ghost-YOLOv8 vehicle detection model is proposed. GhostConv is used to replace part of Conv and a new C2fGhost module is designed to replace part of C2f, reducing the number of parameters and improving the detection performance of the model; The global attention mechanism module is added to the neck network, which strengthens the semantic and location information in the features and improves the model’s ability of feature fusion; In view of the loss of semantic information caused by different scales in detecting small targets, a small target detection layer is added to enhance a combination of deep and shallow semantic information; The GIoU bounding loss function is used instead of the original loss function, which improves the performance of the network’s bounding box regression. Then, combining DeepSORT tracking algorithm and virtual line counting method, the proposed framework estimates traffic flow parameters, including volume, speed, and density in complex environments. The experimental results show that compared with the original model in the UA-DETRAC dataset, the improved YOLOv8 model’s the accuracy is increased by 1.4%, and the parameter number and model size are reduced by 0.229G and 0.2MB respectively. In addition, the framework has good robustness, reaching 97.56% accuracy when estimating average traffic flow parameters. It is sufficient to overcome complex weather conditions and effectively help control traffic for intelligent transportation systems and traffic management to provide good data support. In conclusion, it shows that the model can reduce the number and size of model parameters and improve the detection precision as well as meeting the requirements of edge computing devices and having better real-time performance, so it has practical application value.

Research on Microscale Vehicle Logo Detection Based on Real-Time DEtection TRansformer (RT-DETR)

Vehicle logo detection (VLD) is a critical component of intelligent transportation systems (ITS), particularly for vehicle identification and management in dynamic traffic environments. However, traditional object detection methods are often constrained by image resolution, with vehicle logos in existing datasets typically measuring 32 × 32 pixels. In real-world scenarios, the actual pixel size of vehicle logos is significantly smaller, making it challenging to achieve precise recognition in complex environments. To address this issue, we propose a microscale vehicle logo dataset (VLD-Micro) that improves the detection of distant vehicle logos. Building upon the RT-DETR algorithm, we propose a lightweight vehicle logo detection algorithm for long-range vehicle logos. Our approach enhances both the backbone and the neck network. The backbone employs ResNet-34, combined with Squeeze-and-Excitation Networks (SENetV2) and Context Guided (CG) Blocks, to improve shallow feature extraction and global information capture. The neck network employs a Slim-Neck architecture, incorporating an ADown module to replace traditional downsampling convolutions. Experimental results on the VLD-Micro dataset show that, compared to the original model, our approach reduces the number of parameters by approximately 37.6%, increases the average accuracy (mAP@50:95) by 1.5%, and decreases FLOPS by 36.7%. Our lightweight network significantly improves real-time detection performance while maintaining high accuracy in vehicle logo detection.

Degradation Diagnosis and Control Strategy for a Diesel Hybrid Powertrain Considering State of Health

Hybrid electric vehicles (HEV) are a practical choice for energy saving in the transportation field. Degradation diagnosis (DD) is one of the main methods to guarantee system robustness. However, the classical DD methods cannot meet the requirements of HEV due to their system complexity. In this study, a novel Prognostics and Health Management (PHM) study was conducted to face these challenges. Firstly, a physical P2 HEV model with a rule-based controller was built, and its diesel engine sub-model was simplified by a neural network (NN) to ensure real-time performance of the degradation prognostics. Secondly, a degradation prognostics method based on gray relation analysis–principal component analysis (GRA-PCA) was illustrated, which could confirm degradation 2 s after the health index fell below the threshold. Finally, a degradation tolerance strategy based on long short term memory–model predictive control (LSTM-MPC) was performed to optimize vehicle speed tracing with minimal energy consumption and was validated by three cases. The result shows that the energy consumption stayed nearly unchanged for the engine degradation case. For the battery degradation case, the tracing error was reduced by 11.7% with 4.3% more energy consumption. For combined degradation, the strategy achieved a 12.3% tracing error reduction with 3.7% more energy consumption. The suggested PHM method guaranteed vehicle power performance under degradation situations.

Application of Artificial Intelligence Technology in New Energy Vehicles

Due to environmental pollution caused by excessive exploitation of world resources, countries are now advocating for green travel, which has led to the rapid development of the new energy vehicle industry. Compared to traditional cars, the safety, maneuverability, and convenience of new energy vehicles have made the application of artificial intelligence technology increasingly widespread in this field. Therefore, this article provides a systematic review and analysis of the application of artificial intelligence technology in new energy vehicles. Firstly, the article outlined the main application scenarios of artificial intelligence in the field of new energy vehicles, including autonomous driving, energy management optimization, vehicle intelligent interconnection, and predictive maintenance. On this basis, this paper analyzed the role of artificial intelligence technology in improving the performance, safety, and reliability of new energy vehicles, as well as the technological challenges and development trends it faces. For example, issues such as data privacy protection and real-time performance need to be further addressed. In the future, artificial intelligence and new energy vehicles will further integrate, making important contributions to achieving intelligent, green, and safe future travel.

A Small-Scale Object Detection Algorithm in Intelligent Transportation Scenarios

In response to the problem of poor detection ability of object detection models for small-scale targets in intelligent transportation scenarios, a fusion method is proposed to enhance the features of small-scale targets, starting from feature utilization and fusion methods. The algorithm is based on the YOLOv4 tiny framework and enhances the utilization of shallow and mid-level features on the basis of Feature Pyramid Network (FPN), improving the detection accuracy of small and medium-sized targets. In view of the problem that the background of the intelligent traffic scene image is cluttered, and there is more redundant information, the Convolutional Block Attention Module (CBAM) is used to improve the attention of the model to the traffic target. To address the problem of data imbalance and prior bounding box adaptation in custom traffic data sets that expand traffic images in COCO and VOC, we propose a Copy-Paste method with an improved generation method and a K-means algorithm with improved distance measurement to enhance the model’s detection ability for corresponding categories. Comparative experiments were conducted on a customized 260-thousand traffic data set containing public traffic images, and the results showed that compared to YOLOv4 tiny, the proposed algorithm improved mAP by 4.9% while still ensuring the real-time performance of the model.

Predicting Shot Accuracy in Badminton Using Quiet Eye Metrics and Neural Networks

This paper presents a novel approach to predicting shot accuracy in badminton by analyzing Quiet Eye (QE) metrics such as QE duration, fixation points, and gaze dynamics. We develop a neural network model that combines visual data from eye-tracking devices with biomechanical data such as body posture and shuttlecock trajectory. Our model is designed to predict shot accuracy, providing insights into the role of QE in performance. The study involved 30 badminton players of varying skill levels from the Chinese Swimming Club in Singapore. Using a combination of eye-tracking technology and motion capture systems, we collected data on QE metrics and biomechanical factors during a series of badminton shots for a total of 750. Key results include: (1) The neural network model achieved 85% accuracy in predicting shot outcomes, demonstrating the potential of integrating QE metrics with biomechanical data. (2) QE duration and onset were identified as the most significant predictors of shot accuracy, followed by racket speed and wrist angle at impact. (3) Elite players exhibited significantly longer QE durations (M = 289.5 ms) compared to intermediate (M = 213.7 ms) and novice players (M = 168.3 ms). (4) A strong positive correlation (r = 0.72) was found between QE duration and shot accuracy across all skill levels. These findings have important implications for badminton training and performance evaluation. The study suggests that QE-based training programs could significantly enhance players’ shot accuracy. Furthermore, the predictive model developed in this study offers a framework for real-time performance analysis and personalized training regimens in badminton. By bridging cognitive neuroscience and sports performance through advanced data analytics, this research paves the way for more sophisticated, individualized training approaches in badminton and potentially other fast-paced sports. Future research directions include exploring the temporal dynamics of QE during matches and developing real-time feedback systems based on QE metrics.

Imprecision and real-time clinical performance of a whole blood high sensitivity point of care troponin i: ready for prime time?

Abstract Introduction High sensitivity troponin (hs-cTn) measurement is central to rapid diagnostic algorithms in acute coronary syndrome (ACS). Point of care (POC) testing reduces turnaround time when compared to central lab (CL) testing. Currently there is very little data on the performance of POC hs-cTn assays in a clinical environment with non-expert operators. Method In a nested sub-study of a randomised controlled trial (RCT), patients with suspected ACS were randomised to either the ESC 0/1 or 0/3 hour algorithm. They underwent whole blood (WB) sampling for Siemens Atellica VTLi hs-cTnI POC (VTLi POC), CL hs-cTnI and CL hs-cTnT assays. This was for both initial and repeat samples. Assay imprecision of the VTLi POC assay was assessed by 2 runs of 10 tests with whole blood at 9 different troponin levels. 4 devices were used with samples stored at room temperature.Final diagnosis was adjudicated based on all relevant clinical and imaging data together with CL hs-cTnT as the diagnostic biomarker in clinical use (10% coefficient of variation (CV) 3-5 ng/L, 99th percentile 14 ng/L) and using the 4th universal definition of myocardial infarction (MI). Clinical performance was assessed using Receiver Operator Characteristics (ROC) curves. Results 2144 patients consented and had WB VTLi POC, CL hs-cTnI and CL hs-cTnT available. The index adjudicated type 1 MI rate was 6.1%. Assay imprecision of the VTLi POC assay demonstrated a 10% Coefficient of Variation (CV) at 9.8ng/l. This was >50% lower than the 99th centile of 23ng/l (average for males and females) (Figure 1). Clinical performance of the VTLi POC assay was acceptable and equivalent to CL hs-cTnT: Area Under Curve (AUC) 0.92 (0.89-0.96), 0.98 (0.96-0.99) and 0.93 (0.87-0.98) at presentation (Figure 2), 1 and 3 hours respectively. However, clinical performance was inferior to CL hs-cTnI: AUC 0.96 (0.95-0.98) and 0.96 (0.93-0.98) at presentation (p=0.0007) and 3 hours (p<0.0001) respectively. Conclusion The Siemens Altellica VTLi hs-cTnI POC assay has imprecision levels consistent with a high sensitivity troponin assay. It has acceptable clinical performance though inferior to the equivalent CL hs-cTnI. Further validation of this assay, particularly with optimised single sample rule-out, could facilitate its use in routine clinical practice.

Path tracking with switchable stability constraints for vehicle collision avoidance based on extended stability region

There is a certain conflict between vehicle stability and the performance of path-tracking under emergency conditions. In this paper, to make full use of the lateral capacity of vehicles, a path tracking method with switchable stability constraints is proposed. First, using the Lyapunov energy equation and Hurwitz’s theorem, the lateral stability region (SR) and the critical steering angle of steady steering at different longitudinal speeds are calculated. Based on this, the extended stability region (ESR) and the extended critical steering angle of the vehicle under the control of an additional yaw moment are obtained. Furthermore, the driving area during collision avoidance is divided into unsafe and safe areas, and based on the model predictive control strategy, a path-tracking controller with switchable stability constraints is established. In the unsafe area, the vehicle states are allowed to break through the SR temporarily, and to ensure path-tracking accuracy, the ESR is taken as the stability constraint. In the safe area, the stability constraint is switched to the SR, and to make the vehicle states quickly return to SR, a regain stability controller is established. Finally, a Simulink-CarSim simulation and a hardware-in-the-loop (HIL) experiment are conducted. The simulation results show that a balance between tracking effect and vehicle stability can be effectively achieved through the proposed path-tracking system under the investigated emergency conditions. The real-time performance of the proposed control strategy is validated through the HIL experiment.

Improved A-STAR Algorithm for Power Line Inspection UAV Path Planning

The operational areas for unmanned aerial vehicles (UAVs) used in power line inspection are highly complex; thus, the best path planning under known obstacles is of significant research value for UAVs. This paper establishes a three-dimensional spatial environment based on the gridding and filling of two-dimensional maps, simulates a variety of obstacles, and proposes a new optimization algorithm based on the A-STAR algorithm, considering the unique dynamics and control characteristics of quadcopter UAVs. By utilizing a novel heuristic evaluation function and uniformly applied quadratic B-spline curve smoothing, the planned path is optimized to better suit UAV inspection scenarios. Compared to the traditional A-STAR algorithm, this method offers improved real-time performance and global optimal solution-solving capabilities and is capable of planning safer and more realistic flight paths based on the operational characteristics of quadcopter UAVs in mountainous environments for power line inspection.

Implementing Real-Time Image Processing for Radish Disease Detection Using Hybrid Attention Mechanisms

This paper developed a radish disease detection system based on a hybrid attention mechanism, significantly enhancing the precision and real-time performance in identifying disease characteristics. By integrating spatial and channel attentions, this system demonstrated superior performance across numerous metrics, particularly achieving 93% precision and 91% accuracy in detecting radish virus disease, outperforming existing technologies. Additionally, the introduction of the hybrid attention mechanism proved its superiority in ablation experiments, showing higher performance compared to standard self-attention and the convolutional block attention module. The study also introduced a hybrid loss function that combines cross-entropy loss and Dice loss, effectively addressing the issue of class imbalance and further enhancing the detection capability for rare diseases. These experimental results not only validate the effectiveness of the proposed method, but also provide robust technical support for the rapid and accurate detection of radish diseases, demonstrating its vast potential in agricultural applications. Future research will continue to optimize the model structure and computational efficiency to accommodate a broader range of agricultural disease detection needs.

Fringe photometric stereo

In fringe projection profilometry (FPP), a high spatial frequency of fringe typically indicates powerful error suppression capability; however, it complicates phase unwrapping, necessitating a greater number of patterns and thus compromising the real-time performance of scanning. In this letter, we report a fringe photometric stereo (FPS) to completely bypass phase unwrapping for recovering the continuous 3D surface. We reveal the linear mapping relationship between the phase gradient and depth gradient, thereby facilitating the computation of depth gradients from phase gradients. Thus, we can employ depth gradients to reconstruct high-quality 3D profiles. Experimental results demonstrate that our FPS surpasses traditional phase-shifting profilometry in mitigating Gaussian noise. Our approach enables high-quality and unambiguous 3D reconstruction using single-frequency fringe patterns, relying solely on a camera and a projector without the need for dual, multiple, or light field cameras, thereby paving a path to high-speed and precision 3D imaging.

LIVS: a tightly coupled LiDAR, IMU, and camera-based SLAM system under scene degradation

In increasingly complex environments, single-sensor-based SLAM systems face scene degradation problems such as strong light, dynamic obstacle interference, and lack of structural features in the environment, which severely constrain the localization accuracy and mapping effectiveness of SLAM systems. To address this problem, this paper proposes a tightly coupled laser-IMU-visual SLAM system (LIVS) based on factor graph optimization. LIVS consists of two subsystems: the laser inertial system (LIS) and the visual inertial system (VIS). The pose estimation of the LIS can provide an initialization for the VIS; LiDAR suffers scene degradation in environments lacking structural features, while the VIS provides better constraints for Lidar in this environment. Firstly, a method of selecting keyframes based on ORB features and the relative motion of the mobile robot is proposed as a means of improving the positioning accuracy and tracking ability of the system in the case of strenuous motion, and a strategy of local sliding window optimization is adopted to effectively improve the real-time performance and robustness of the system. Meanwhile, in order to further improve the global map consistency and positioning accuracy, a hybrid closed-loop detection method based on laser and vision is proposed, which utilizes the geometrical features of laser and the image features of vision to accomplish the hybrid closed-loop detection. Secondly, the laser inertia factor, IMU pre-integration factor, vision inertia factor, and closed-loop factor are globally optimized in a factor map framework to achieve accurate estimation of the robot pose. Finally, this paper presents experimental evaluations on both public and homemade datasets and compares the performance of the proposed algorithm with that of other algorithms. Experimental results show that the algorithm in this paper exhibits better performance in both localization accuracy and mapping effectiveness.

CMDN: Pre-Trained Visual Representations Boost Adversarial Robustness for UAV Tracking

Visual object tracking is widely adopted to unmanned aerial vehicle (UAV)-related applications, which demand reliable tracking precision and real-time performance. However, UAV trackers are highly susceptible to adversarial attacks, while research on developing effective adversarial defense methods for UAV tracking remains limited. To tackle these challenges, we propose CMDN, a novel pre-processing defense network that effectively purifies adversarial perturbations by reconstructing video frames. This network learns robust visual representations from video frames, guided by meaningful features from both the search region and the template. Comprehensive experiments on three benchmarks demonstrate that CMDN is capable of enhancing a UAV tracker’s adversarial robustness in both adaptive and non-adaptive attack scenarios. In addition, CMDN maintains stable defense effectiveness when transferred to heterogeneous trackers. Real-world tests on the UAV platform also validate its reliable defense effectiveness and real-time performance, with CMDN achieving 27 FPS on NVIDIA Jetson Orin 16 GB (25 W mode).

Adaptive optical correction in in vivo two-photon fluorescence microscopy with neural fields.

Adaptive optics (AO) techniques are designed to restore ideal imaging performance by measuring and correcting aberrations originating from both the microscope system and the sample itself. Conventional AO methods require additional hardware, such as wavefront sensors and corrective devices, for aberration measurement and correction, respectively. These methods often necessitate microscopes to adhere to strict design parameters, like perfect optical conjugation, to ensure the accurate delivery of corrective patterns for wavefront correction using corrective devices. However, in general microscope systems, including commercially available ones, conjugation errors are more prone to arise due to incomplete conjugation among optical components by design and misalignment of the components, coupled with their limited access and adjustability, which hinders the rigorous integration of AO hardware. Here, we describe a general-purpose AO framework using neural fields, NeAT, that is applicable to both custom-built and commercial two-photon fluorescence microscopes and demonstrate its performance in various in vivo imaging settings. This framework estimates wavefront aberration from a single 3D two-photon fluorescence image stack, without requiring external datasets for training. Additionally, it addresses the issue of incomplete optical conjugation by estimating and correcting any conjugation errors, which enables more accurate aberration correction by the corrective device. Finally, it jointly recovers the sample's 3D structural information during the learning process, potentially eliminating the need for hardware-based AO correction. We first carefully assess its aberration estimation performance using a custom-built two-photon fluorescence microscope equipped with a wavefront sensor which provides the ground truth aberration for comparison. We further characterize and assess the robustness of the aberration estimation to image stacks with low signal-to-noise ratios, strong aberration, and motion artifacts. As practical applications, using a commercial microscope with a spatial light modulator, we first demonstrate NeAT's real-time aberration correction performance in in vivo morphological imaging of the mouse brain. We further show its performance in in vivo functional activity imaging of glutamate and calcium dynamics within the mouse brain.

Three-dimensional finite element analysis of biological signal feedback in the mechanical properties of sound production

In response to the problems of low reliability and long time consumption in traditional research on sound production, this article used electromyography (EMG) signals as input signals to build a high-precision sound production mechanics model. A Proportional-Integral-Derivative (PID) controller was used to dynamically adjust the model and develop a real-time feedback system. This article established a detailed three-dimensional (3D) finite element model including the vocal cords and throat, defined the nonlinear elastic properties and linear elastic properties of different tissues, and used tetrahedral grid partitioning technology to improve the computational accuracy of the model. Through a EMG sensor, an individual EMG was collected and filtered to remove noise. The processed EMGs were used as input parameters for the finite element model to drive the muscle units in the model. By using a PID controller to receive real-time EMG input, the error was calculated and the model was adjusted, enabling accurate simulation of the mechanical properties of vocal cord vibration under different vocal states and achieving real-time feedback. Considering the complexity of vocal cord vibration driven by biological signals, this article simulated and analyzed the modal characteristics of vocal cord vibration, and analyzed the differences in vocal cord vibration characteristics under different vocal states. The sound pressure distribution and resonance frequency were simulated and analyzed to understand the propagation characteristics of sound. Finally, by comparing and analyzing the simulated data with the actual collected data, it was found that the maximum relative error rate of the model under different sound states was 6.14%, and the overall error rate of the model was relatively low, which verified the reliability of the model. The findings demonstrated that the feedback delays of the model in different sound states were all within 100 milliseconds, indicating that the system had high real-time performance and accuracy, which was promising in practical applications.

Robust Real-Time Cancer Tracking via Dual-Panel X-Ray Images for Precision Radiotherapy

Respiratory-induced tumor motion presents a critical challenge in lung cancer radiotherapy, potentially impacting treatment precision and efficacy. This study introduces an innovative, deep learning-based approach for real-time, markerless lung tumor tracking utilizing orthogonal X-ray projection images. It incorporates three key components: (1) a sophisticated data augmentation technique combining a hybrid deformable model with 3D thin-plate spline transformation, (2) a state-of-the-art Transformer-based segmentation network for precise tumor boundary delineation, and (3) a CNN regression network for accurate 3D tumor position estimation. We rigorously evaluated this approach using both patient data from The Cancer Imaging Archive and dynamic thorax phantom data, assessing performance across various noise levels and comparing it with current leading algorithms. For TCIA patient data, the average DSC and HD95 values were 0.9789 and 1.8423 mm, respectively, with an average centroid localization deviation of 0.5441 mm. On CIRS phantoms, DSCs were 0.9671 (large tumor) and 0.9438 (small tumor) with corresponding HD95 values of 1.8178 mm and 1.9679 mm. The 3D centroid localization accuracy was consistently below 0.33 mm. The processing time averaged 90 ms/frame. Even under high noise conditions (S2 = 25), errors for all data remained within 1 mm with tracking success rates mostly at 100%. In conclusion, the proposed markerless tracking method demonstrates superior accuracy, noise robustness, and real-time performance for lung tumor localization during radiotherapy. Its potential to enhance treatment precision, especially for small tumors, represents a significant step toward improving radiotherapy efficacy and personalizing cancer treatment.

Applications of Machine Learning in Real-Time Control Systems: A Review

Abstract Real-time control systems(RTCS) have become an indispensable part of modern industry, finding widespread applications in fields such as robotics, intelligent manufacturing and transportation. However, these systems face significant challenges, including complex nonlinear dynamics, uncertainties and various constraints. These challenges result in weakened disturbance rejection and reduced adaptability, which make it difficult to meet increasingly stringent performance requirements. In fact, RTCS generate a large amount of data, which presents an important opportunity to enhance control effectiveness. Machine learning, with its efficiency in extracting valuable information from big data, holds significant potential for applications in RTCS. Exploring the applications of machine learning in RTCS is of great importance for guiding scientific research and industrial production. This paper first analyzes the challenges currently faced by RTCS, elucidating the motivation for integrating machine learning into these systems. Subsequently, it discusses the applications of machine learning in RTCS from various aspects, including system identification, controller design and optimization, fault diagnosis and tolerance, and perception. The research indicates that data-driven machine learning methods exhibit significant advantages in addressing the multivariable coupling characteristics of complex nonlinear systems, as well as the uncertainties arising from environmental disturbances and faults, thereby effectively enhancing the system's flexibility and robustness. However, compared to traditional methods, the applications of machine learning also faces issues such as poor model interpretability, high computational requirements leading to insufficient real-time performance, and a strong dependency on high-quality data. This paper discusses these challenges and proposes potential future research directions.

Efficient Nonlinear Model Predictive Path Tracking Control for Autonomous Vehicle: Investigating the Effects of Vehicle Dynamics Stiffness

Motion control is one of the three core modules of autonomous driving, and nonlinear model predictive control (NMPC) has recently attracted widespread attention in the field of motion control. Vehicle dynamics equations, as a widely used model, have a significant impact on the solution efficiency of NMPC due to their stiffness. This paper first theoretically analyzes the limitations on the discretized time step caused by the stiffness of the vehicle dynamics model equations when using existing common numerical methods to solve NMPC, thereby revealing the reasons for the low computational efficiency of NMPC. Then, an A-stable controller based on the finite element orthogonal collocation method is proposed, which greatly expands the stable domain range of the numerical solution process of NMPC, thus achieving the purpose of relaxing the discretized time step restrictions and improving the real-time performance of NMPC. Finally, through CarSim 8.0/Simulink 2021a co-simulation, it is verified that the vehicle dynamics model equations are with great stiffness when the vehicle speed is low, and the proposed controller can enhance the real-time performance of NMPC. As the vehicle speed increases, the stiffness of the vehicle dynamics model equation decreases. In addition to the superior capability in addressing the integration stability issues arising from the stiffness nature of the vehicle dynamics equations, the proposed NMPC controller also demonstrates higher accuracy across a broad range of vehicle speeds.

Driver Distraction Detection Based on Fusion Enhancement and Global Saliency Optimization

Driver distraction detection not only effectively prevents traffic accidents but also promotes the development of intelligent transportation systems. In recent years, thanks to the powerful feature learning capabilities of deep learning algorithms, driver distraction detection methods based on deep learning have increased significantly. However, for resource-constrained onboard devices, real-time lightweight models are crucial. Most existing methods tend to focus solely on lightweight model design, neglecting the loss in detection performance for small targets. To achieve a balance between detection accuracy and network lightweighting, this paper proposes a driver distraction detection method that combines enhancement and global saliency optimization. The method mainly consists of three modules: context fusion enhancement module (CFEM), channel optimization feedback module (COFM), and channel saliency distillation module (CSDM). In the CFEM module, one-dimensional convolution is used to capture information between distant pixels, and an injection mechanism is adopted to further integrate high-level semantic information with low-level detail information, enhancing feature fusion capabilities. The COFM module incorporates a feedback mechanism to consider the impact of inter-layer and intra-layer channel relationships on model compression performance, achieving joint pruning of global channels. The CSDM module guides the student network to learn the salient feature information from the teacher network, effectively balancing the model’s real-time performance and accuracy. Experimental results show that this method outperforms the state-of-the-art methods in driver distraction detection tasks, demonstrating good performance and potential application prospects.