As the core carrier of intelligent interaction， innovative breakthroughs in digital human technology are of crucial significance to the in-depth integration of artificial intelligence and the digital economy. Focusing on the field of computer vision， this paper discusses the two core themes of digital human technology—modeling and driving—and systematically sorts out the characteristics of relevant datasets and evaluation methods. In terms of modeling， the research starts with traditional geometric modeling techniques， covering refined three-dimensional reconstruction methods based on mesh optimization and point cloud processing， and further explores new paradigms of generative modeling driven by deep learning. In terms of driving， it focuses on analyzing human pose estimation and facial expression transfer techniques based on video sequences， as well as speech-driven lip-sync generation algorithms combined with audio features. The scale and diversity of datasets are vital to the generation of digital human appearances， while the sophistication of evaluation methods allows for a more objective measurement of generation performance. This paper systematically categorizes and summarizes representative works in the field of digital human modeling and driving， analyzes the advantages and limitations of existing methods， and prospects potential future research directions in combination with current technological development trends.

To understand the research status and future research hotspots in the field of vehicle-infrastructure integration at intersections， the Web of Science and CNKI databases were used as data sources to retrieve core journal articles from January 2000 to September 2024， totaling 427 articles. Through bibliometric analysis， statistics and visualization were performed from the perspectives of publication trends， literature sources， and keyword co-word analysis. The statistical results show that the number of publications in this field has been increasing in recent years， and the sample documents are highly representative. Keyword co-occurrence analysis shows that the research directions in this field mainly include intersection control， path and speed guidance， signal-vehicle collaborative optimization， modeling and simulation， multi-vehicle interaction safety， single vehicle safety， vulnerable road user safety， network communication， and roadside perception. The development context and specific methods of the above research directions were sorted out and summarized， and based on this， the future research hotspots in the field of intersection vehicle-infrastructure integration were prospected.

A one-dimensional transient model of a China VI diesel engine equipped with a diesel particulate filter （DPF） was developed using GT-Power. This study explores the effects of DPF on diesel engine transient characteristics under varying altitudes， specifically focusing on constant-speed， varying-torque scenarios. It also examines how changes in ash layer permeability impact engine transient performance and DPF operation， while optimizing the DPF substrate ratio and diameter range. The results show that the DPF reduces transient intake flow， torque， and thermal efficiency compared to the original engine， while increasing fuel consumption， particularly at high altitudes and in the later stages of constant-speed， increasing-torque conditions. Calcium-based ash has the most significant impact on transient characteristics. During the later stages of these conditions under varying altitudes， magnesium- and zinc-based ash cause a shift in DPF pressure drop. When the substrate ratio exceeds 1.2 and the substrate diameter exceeds 190 mm， the impact on additional fuel consumption stabilizes， although increased carbon loading intensifies this trend.

In the vector control system of a T-type three-level inverter for permanent magnet synchronous motors，switch failures such as open circuits or short circuits are common occurrences.These failures can disrupt the original modulation strategy，leading to distortion in the three-phase currents，which in turn causes fluctuations in speed and torque，severely impacting system stability.To mitigate these issues，this study proposes a fault-tolerant topology for the T-type three-level inverter and a corresponding fault-tolerant five-segment space vector pulse width modulation（SVPWM）strategy.The new topology can uniformly convert potential short-circuit faults into open-circuit faults，thereby reducing the damage caused by short-circuit faults to the system.Faults are categorized based on the number of affected phases，and when the fault type meets the feasibility requirements of the fault-tolerant modulation strategy，the system automatically switches to the fault-tolerant five-segment SVPWM strategy to ensure normal operation during fault conditions.Simulation and experimental results validate the effectiveness and reliability of the proposed topology and modulation strategy in practical applications.

Accurate and real-time evaluation of the battery's state of health is the core of the battery management system in electric vehicles. This paper proposes a novel model for estimating the full charging time of lithium batteries. Firstly， utilizing the high estimation accuracy of the Unscented Kalman Filter for nonlinear problems， a dual Unscented Kalman Filter prediction correction framework with further improved accuracy is designed， which can accurately estimate the current full charge time of lithium batteries. Under this framework， the measurement equation of the Unscented Kalman Filter is linearly weighted using Gaussian Process Regression and Support Vector Regression prediction results. The experimental results show that the framework proposes in this paper has high accuracy and real-time performance， with an average relative error of 0.001 6 for estimating 180 full charge times. Compared with the EKF and DEKF based algorithms， the average relative error has reduced by 98.87% and 98.15%， respectively.

In order to improve the tracking accuracy and anti-interference ability of the permanent magnet synchronous motor in the coarse tracking system of space laser communication platform， a robust adaptive backstepping controller based on improved disturbance observer was designed.Firstly， the backstepping controller based on disturbance observer is designed to deal with the uncertainty of external disturbance， and the disturbance observer is improved， which has gotten a very good effect. Then， an adaptive backstepping controller based on a disturbance observer is designed to deal with the time-varying internal parameters of the permanent magnet synchronous motor within a certain range during actual operation， which significantly reduces the system tracking error.Finally， in order to further reduce the disturbance of uncertain factors such as observation error in the running process of the system， a robust feedback item is added to the adaptive backstepping controller， and the tracking accuracy of the system is improved again.Results in simulation show that the proposed controller is able to effectively reduce the position tracking error and has a good disturbance rejection performance.

When the vibration frequency of the three cylinder fracturing pump coincides with the natural mode frequency during operation， the resonance phenomenon generated will intensify the vibration of the fracturing pump， causing severe fluctuations in the amplitude of the vibration displacement， thereby reducing the stability of the three cylinder fracturing pump operation. Therefore， a vibration frequency control algorithm for three cylinder fracturing pumps considering excitation peak values is proposed. By considering the combined effects of the natural modal frequency， namely the inertia force of the crank slider mechanism， the hydraulic wave force at the hydraulic end， and the frictional force， the excitation peak value of the three cylinder fracturing pump is determined. Using the FBLMS control algorithm， the error signal between the peak excitation value and the expected target value of the three cylinder fracturing pump is transformed in the frequency domain to obtain the response result of the vibration frequency. Then， FFT technology is used to determine the position of the vibration signal， and the vibration acceleration weight is updated based on the signal position to reduce the vibration displacement amplitude and achieve vibration frequency control of the three cylinder fracturing pump. The experimental results show that the algorithm effectively reduces the vibration amplitude of the three cylinder fracturing pump to about 0.1 μm， and improves the stability of the three cylinder fracturing pump operation.

In order to further improve the safety of truck braking process， the reliability of truck stopping sight distance on downhill section of highway is studied. A four-stage stopping sight distance model considering the ABS is constructed. The reliability theory is introduced to analyze the reliability of truck stopping sight distance on the downhill section of the highway by using Monte Carlo method， and the suggested value of truck stopping sight distance is given. Furthermore， the rationality of the model is verified by engineering examples and TruckSim simulation. The results show that there is a positive correlation between the probability of stopping sight distance failure and the accident rate； the error of the calculated value of the reliability model is smaller than the standardized value， and the results are more conservative. It shows that the research results have certain practical significance for reducing road safety risks.

To study the following characteristics and stability conditions of vehicles on curved roads， this paper builds upon an intelligent driver model （IDM）， integrating considerations of factors such as curve radius and super-elevation， and incorporating road mechanics to establish an extended model. By analyzing the influence of curves on traffic flow stability， the study derives stability conditions applicable to both homogeneous and heterogeneous traffic flows comprising human driven vehicles （HDVs）， automated vehicles （AVs）， and connected automated vehicles （CAVs）. Parameter sensitivity analyses and numerical validations are conducted. The findings indicate that traffic flow stability is highest under Condition 2 （R=\mathrm{250}\text{?}\mathrm{m},i_{\mathrm{e}}=\mathrm{8}\mathrm{\%}）. Increasing equilibrium speeds enhances traffic flow stability， while increasing free-flow speeds is detrimental to stability. Moreover， as CAVs penetration rates increase， the stability of mixed traffic flows gradually improves， although increasing maximum platoon sizes may destabilize traffic flows. In homogeneous traffic flows， lower speeds are advantageous for stability at low speeds； for high-speed conditions， stability is optimized at speeds of 29.9 m/s for HDVs， 26.8 m/s for AVs， and 26.9 m/s for CAVs. In mixed traffic flows， stability can be maintained when the penetration rates of CAVs exceed 0.72.

Considering the economy of the damper and the replaceability after the earthquake， the friction damper with replaceable rubber in steel pipe （SPRFD） composed of high-strength bolts， steel pipe， movable baffles， fixed baffles， steel tie rods， and replaceable rubbers was proposed， and its basic structure and working principle were introduced. Through reciprocating loading test， the change rule of the hysteresis curve of the damper under different preloads were investigated， the change rule of the hysteresis response characteristics， equivalent stiffness and equivalent damping ratios under different loading displacements were analyzed， and the friction damage characteristic of the replaceable rubber was obtained. The seismic performance of railway self-centering bridge pier equipped with SPRFD was investigated through cyclic loading analyses. The results showed that the hysteresis performance of the damper was stable under the smaller preload， and the hysteresis loop had the characteristics of rigid-plastic model； As the preload increasing， the reciprocating cycle and large displacement loading decreased the damping force causing by rubber damage， and the hysteresis loop had the characteristics of elastic-plastic model. During the whole loading process， the dampers had a relatively stable equivalent viscous damping ratio. SPRFD with different preload had different effects on the hysteresis curve of the self-centering bridge piers. The value of the preload affected the unloading stiffness of hysteresis curve. The hysteresis curve had an obvious pinching under the larger preload. The larger preload helped the self-centering bridge pier enhance lateral strength.

To address the limitations of traditional models in accurately characterizing the dynamic evolution of vehicle platoons and interactions with heterogeneous behaviors of human-driven vehicles （HDV） in mixed traffic flows， this paper proposes a hybrid traffic flow modeling framework integrating discrete motion rules and dynamic platoon evolution. First， a discrete motion safety distance model is developed to resolve the distortion problem in continuous acceleration modeling by introducing integer decision rules based on cellular spacing. Second， a platoon size transition probability matrix is constructed using Markov chains to dynamically characterize the splitting， merging， and reorganization processes of platoons. Finally， multi-scenario simulation data are employed to quantify the impact mechanisms of intra-platoon spacing， reaction time， lane-changing behavior， and platoon size on CO? emissions. The results indicate that： When the CAV penetration rate exceeds 0.6， platoon mode can effectively improve the operational state of mixed traffic flow， and significantly reduces CO? emissions， with a reduction range of 18.2% to 25.1%； Optimizing intra-platoon spacing achieves a peak emission reduction of 42.5%； Lane-changing strategies must adapt to traffic density—enhancing HDV lane-changing probability under low density can achieve carbon emission reductions， while restricting CAV lane-changing probability to 0.4~0.6 under high density； Platoon sizes of 3~5 vehicles demonstrate optimal emission reduction efficiency. This paper reveals the mechanism of queue dynamic parameters on carbon emissions， and provides theoretical support for the optimization of intelligent connected fleet cooperative control strategies and the design of low-carbon transportation systems.

This study focuses on the overtaking lane-changing behavior in urban expressway interchange weaving areas， aiming to analyze the lane-change space selection characteristics of overtaking vehicles and explore control methods for overtaking lane-change behavior. Utilizing real-time trajectory data， this study analyzes the differences in gap selection and lane-changing point selection across various stages of the overtaking process. Machine learning methods were used to predict lane-change duration and lane-change space selection changes. Based on the prediction results， a speed optimization control model for overtaking lane-change behavior was established. The control effect of the model was then tested using a cellular automata simulation environment. The results show that under the control model， the proportion of vehicles selecting "excellent" and "good" grade lane-change gaps and the optimal lane-change point position increased by up to 18.86 and 6.89 percentage points compared to actual values. Additionally， the operating speeds of the three lanes in the weaving area increased by 6.91%， 1.71%， and 3.85%， respectively， and the spatiotemporal utilization rates of the lanes also exhibited better balance.

To address the limitations of classical path planning algorithms， which often overlook multi-modal transportation combinations and lack modeling of waiting times， a study was conducted on the transportation network in Xiamen City. The research involved constructing a multi-layer network topology that considers waiting times. Based on the characteristics of FINLO， a time-dependent modeling method for waiting times associated with different travel modes was proposed. Finally， the time-dependent waiting functions were incorporated into a multi-layer network path planning algorithm based on an adjacency dictionary and binary heap Dijkstra's algorithm. The results showed that this approach effectively considers waiting time dependencies related to multi-modal transit， vehicle schedules， and signal control. Compared to path planning algorithms that do not account for waiting time dependencies， the proposed method reduced total travel time by 8.4%.

In order to analyze the influence of travel time correlation and resource constraints on the selection of the most reliable path in traffic network， the model of the constrained most reliable path with stochastic traffic network considering travel time correlation is established， and the solution method of improved alternating direction method of multipliers is proposed. By Cholesky decomposition and quadratic term deflation， the objective function is transformed into a matrix form of separable structure， and the augmented Lagrangian relaxation problem is decomposed into a series of subproblems by the block coordinate descent method， the upper and lower bounds are iterated to obtain the approximate optimal solution. Numerical simulation experiments are carried out for Sioux Falls network and Chicago sketch network， and the performance of improved alternating direction method of multipliers and Lagrangian relaxation algorithm is compared and analyzed. The results show that the most reliable path problem with or without link travel time correlation is different， the travel time correlation and resource constraints have a significant impact on the selection of the most reliable path， the computational efficiency and convergence of the improved alternating direction method of multipliers are superior to the Lagrangian relaxation algorithm.

This paper focuses on the urban last-mile delivery process under the drone service model， optimizing drone delivery facility location and service allocation strategies through a two-stage mathematical programming model. The impact of demand uncertainty on location-allocation decisions is characterized using a distributionally robust optimization method. The original model is equivalently transformed into a mixed-integer second-order cone programming problem， and an outer approximation algorithm is proposed to improve solving efficiency. Simulation verification is conducted based on the pharmaceutical delivery data in Songjiang District， Shanghai. The results indicate that the proposed outer approximation algorithm can reduce the computational time of commercial solvers by 32.07%； compared with the deterministic model， the proposed distributionally robust optimization model can increase the system's total profit by 4.21 times when demand is highly volatile.

The existing research on multimodal driving fatigue detection mainly focuses on the application level， lacking in-depth exploration of the underlying mechanisms between each mode and driving fatigue. This study aims to explore the relationship between commonly used data sources （including electroencephalogram， electrocardiogram， and vehicle motion information parameters） and driving fatigue， and analyze the potential correlation between these data sources in driving fatigue detection. We collected 32 sets of data through driving simulation experiments and evaluated subjective fatigue levels using the Karolinska Sleepiness Scale， with eyelid closure as an indicator of objective fatigue levels. The parameters of the three modalities are the standard deviation of the R-peak interval of the electrocardiogram （RMSSD）， the power ratio of the EEG frequency band （α+θ/β）， and the standard deviation of the vehicle lateral offset （SDLP）. In the exploratory factor analysis results， the variance explained by the first three factors exceeds 50%. A potential relationship model between various modes and driving fatigue was constructed using structural equation modeling. The analysis results showed that each mode can explain the variability of driving fatigue to a certain extent. Among them， RMSSD and SDLP have significant advantages in predicting subjective fatigue， while α+θ/β shows a close correlation with objective fatigue.

The intersection and overlap of urban subway and bus network stations， the complexity of routes， and the tidal phenomenon of passenger flow during peak hours make it difficult for the transportation network with unreasonable layout to complement resources， resulting in longer travel time and increased carbon emissions for passengers transferring between routes. To this end， an improved ant colony algorithm is proposed to optimize the efficiency of urban subway bus coupling network layout. This method completes the complex topology connection of overlapping lines and stations in the urban subway bus coupling network by coupling station pairs and coupling distances， achieving complementary subway bus transportation resources； based on topological structure， design the objective function of transfer station layout to reduce the travel time and transportation carbon emissions of transfer passengers， as well as the constraint conditions to maximize carbon emission benefits， in order to solve the problems of prolonged travel time and increased transportation carbon emissions of transfer passengers； improve the adaptive setting method of pheromone volatilization coefficient for traditional ant colony algorithm， and quickly solve the layout scheme of subway bus coupling network transfer station location and route direction that meets the objective function and constraint conditions. The research results show that this method can associate complex urban subway bus coupling transfer networks with coupling station pairs and coupling lines to complete coupling modeling. After improving the ant colony algorithm， the maximum solution time for layout optimization schemes in this article is less than 1 second， which is significantly lower than before optimization. After optimizing the layout of the urban subway bus coupling network， the change in walking distance for transfer passengers is -16 m， and the walking time is reduced by -5.46%. The total travel time of transfer passengers decreased by 1.18 hours. The coupling network between urban subway and public transportation has improved transfer efficiency， significant carbon emission benefits， and is more efficient in solving layout optimization solutions.

To achieve rapid and accurate detection of various abnormal conditions on complex road surfaces， a road surface anomaly detection algorithm ATFL-YOLOv8 is proposed from a driving perspective. Firstly， the ADown convolution module is used to replace some of the ordinary convolutions in the baseline model （YOLOv8n） for efficient feature extraction and sampling； secondly， a Triplet attention layer is added at the end of the baseline model backbone network to enhance the model's perceptual ability； again， using the idea of partial convolution， a new lightweight module C2f Faster is constructed to replace the C2f module in the neck network of the baseline model； finally， the introduction of a brand new LSCD Head detection head further reduces the number of model parameters and improves model detection performance. The test results show that compared to the baseline model， ATFL-YOLOv8 has mAP0.5 and mAP@0.5 0.95 increased by 3.1% and 4.2% respectively， reaching 89.9% and 59.7%； at the same time， the number of parameters， floating-point operations， and model size decreased by 47%， 37%， and 45%， respectively， to 1.61 M， 5.2 G， and 3.31 MB. Through actual vehicle verification， it has been proven that it has the ability to detect road surface abnormalities at certain low speeds.

This paper proposes a bridge anomaly detection model that integrates temporal and spatial features TSSF-BADM. The model utilizes graph attention networks （GAT） and long short-term memory （LSTM） networks to extract temporal and spatial data features. These multi-dimensional features are then fused using gated recurrent units （GRU） to capture sequential patterns in the time series. The fused data undergo joint optimization through prediction and reconstruction models， employing stacked LSTM networks and variational autoencoders （VAE） for prediction and reconstruction， respectively. Finally， the prediction and reconstruction errors of the model are analysed using the peak over threshold （POT） method to obtain the threshold and perform anomaly detection， and the samples exceeding the anomaly threshold are considered as anomalous samples. The experimental comparison results show that the model in this paper achieves good performance on the real bridge Z24， the accuracy of recognition reaches 0.986 8， and the recognition delay is only 0.008， which are all better than other comparative models such as LSTM_VAE， MAD_GAN， OmniAnomaly， etc.， and are able to effectively carry out the detection of the abnormal state of the bridge. It provides decision-making for bridge safety detection， preventive maintenance， etc.

In imbalanced data classification， the majority of class samples have an advantage in terms of quantity， and their distribution will have a significant "pulling" effect on the clustering results. However， the minority class samples， due to their small quantity， have relatively unclear features in the entire dataset， resulting in drift problems in the data stream and affecting the classification performance of the data stream. To address this issue， research is conducted on an imbalanced drift big data stream classification algorithm based on spectral clustering undersampling. By using undersampling techniques to reduce the redundant amount of majority class data in imbalanced drift big data streams， balance the amount of majority class data and minority class data， and alleviate the problem of data drift caused by clustering "pulling"； select the core points of the balanced big data stream to form a core point set， and use spectral clustering algorithm to cluster this core point set. Based on the clustering structure obtained from spectral clustering and the selected core points， realize the classification of imbalanced drift big data streams. The experimental results show that the algorithm can achieve balanced processing of imbalanced drift big data streams， and the average imbalance degree after processing can be reduced to 1.024， almost approaching the equilibrium state； it can achieve the selection and effective grouping of core points for different attribute big data streams， providing guarantees for the subsequent effective application of such big data streams.

In video captioning tasks， one of the common problems is that the object caption is not specific enough， mainly because the model does not fully learn the information of the objects in the video. Meanwhile， videos contain abundant feature information， such as object information， motion information， and contextual information， making it a challenging task to enhance the model’s ability to learn key information when generating captions. To address the aforementioned problems， this paper proposes a method based on enhanced object learning and attention networks. Firstly， a new enhanced object learning module was designed to fully learn object information in videos， thereby achieving accurate caption of video content. Secondly， an attention network was constructed to dynamically adjust the weights of different types of information， thereby enhancing the model’s ability to learn key information when generating captions. In the experiments on the MSVD and MSR-VTT datasets， the caption generated by the method proposed in this paper showed a higher level of specificity and accuracy， and exceeded the current advanced methods in various evaluation indicators， effectively verifying the feasibility of the method.

To fully leverage the appearance features of RGB images and the geometric features of depth images， this paper proposes an "appearance-geometry" features parallel fusion method for object position estimation. First， in the feature extraction and fusion stage， a three-stream bidirectional fusion architecture is constructed to ensure that the parallel RGB image features and depth image features are fused at each encoding layer and decoding layer. To prevent the loss of important features and achieve sufficient fusion of the two types of features， two complementary attention mechanisms are designed， enabling the two features to gain both local and global complementarities. Sercond， in the pose inference calculation stage， considering the distance between the keypoints output by the network and the object’s center point， a keypoint detection network based on a combination of distance metric and distance constraint is proposed， achieving accurate position estimation. The proposed algorithm has been tested on two challenging 6D object position estimation datasets， validating its effectiveness.

To solve the problem of existing deep learning-based object detection algorithms being unsuitable for onboard deployment due to their complexity， a multi-scale feature enhanced lightweight detection algorithm （MFLDet） for ship recognition tasks in optical remote sensing images is proposed. Firstly， to reduce the algorithm's parameter and computation load， a lightweight network architecture， PG-HGNet， is constructed as the backbone network， and constructs a lightweight cross-scale feature fusion network for interactive feature integration. Secondly， a multi-scale feature enhancement module （MFEM） is designed to accommodate the scale variability of ship targets in remote sensing images， thereby enhancing detection accuracy. Finally， the MPDIoU bounding box loss function is introduced to adjust for cases where the predicted and actual bounding boxes share the same aspect ratio but differ in absolute dimensions. Comparative experiments conducted on the HRSC2016 dataset demonstrate that， compared to the baseline model， MFLDet achieves a 55% reduction in parameters and a 23.5% reduction in computation， with only a 0.2 percentage points decrease in average precision， thus effectively balancing complexity and accuracy. Overall， the proposed method surpasses other comparative methods in terms of both lightweight level and detection precision.

Aimed at the problems of current hardware-based adaptive optics including incomplete wavefront correction， high hardware cost， and complexity of the system structure， this paper proposed a new simple adaptive optics to achieve high resolution imaging of the observed target. It completely abandoned the traditional wavefront detection and correction devices， while used computational imaging technology to directly correct the wavefront aberrations at the image level through deep learning algorithm. Firstly， based on deep learning algorithm， a system focal plane degraded image is used to calculate atmospheric turbulence aberrations once and for all. Then based on the resolved turbulence aberrations， a deconvolution is processed to the degraded image to obtain high-resolution reconstruction of the observed object. The simulation results show that the proposed deep neural network can solve the atmospheric turbulence aberrations with high accuracy and high speed under the configured hardware environment. The quality of the image recovered by the deconvolution strategy based on the solved aberrations is greatly improved compared with the degraded image without correction， and the high-resolution imaging for the proposed simple adaptive optics without hardware is realized

To solve the problem of a large amount of human resource consumption caused by tampering with video content review and screening， a technique for identifying tampered content in videos is proposed in this paper. First， key frames are extracted by calculating inter-frame differences. Then， the locations of tampering are determined through differential analysis between the key frames and the original frames. Subsequently， based on the phenomenon of location association and semantic aggregation， the problems of location dispersion and semantic dispersion are solved by the DBSCAN algorithm， which is caused by the randomness of tampering with location and content structure. Finally， optical character recognition （OCR） technology is applied to decipher the specific content that has been altered. Spatio-temporal position and content identification of tampered video content are achieved by the proposed method， providing a solid technical foundation for video content inspection in fields such as public safety， media， and business.

To address the challenges of radar signal interception and low sorting accuracy in the electronic warfare domain， this paper proposes a new online sorting method for radar signals based on self-supervised data streams. Firstly， a balanced feature fusion method using generative adversarial networks is employed to achieve high-quality feature fusion and sample enhancement with limited samples. Secondly， a self-supervised model based on multi-task meta-learning is constructed to realize online sorting of unlabeled signal streams. Finally， experimental results show that the proposed method achieves an accuracy of 95.82% on a simulated dataset at 4 dB， and it also performs well on real and public datasets， confirming the effectiveness of the proposed method in online sorting of data streams.

In order to reduce the conflict between vehicle and driver in the human-machine collaboration mode and to realize the flexible transition of control authority. Firstly， a simulated car-following experiment was carried out， and a driving risk assessment model was established based on time to collision （TTC）， time to line crossing （TLC）， and time to brake （TTB）. Secondly， a predictive model of driver ability loss was developed using random forest regression （RFR） based on longitudinal acceleration and steering wheel angle. Finally， a control authority allocation strategy was formulated based on the boundary thresholds of driving risk and driver ability loss. The results show that in longitudinal collision events and near-collision events， the model suggests that the moment when the vehicle should intervene is 0.62 seconds and 1.03 seconds earlier than the driver steps on the brake pedal， respectively， with an effective recognition rate of 87%. The study can provide some theoretical assistance for the design of control allocation and transition between the vehicle and the driver in the human-machine collaborative driving.