Industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI SHINNY INSTR CONTROL TECH
- Filing Date
- 2025-12-16
- Publication Date
- 2026-05-29
AI Technical Summary
Existing industrial vehicle identification and unlocking systems lack personalized control strategies for drivers during vehicle operation. They cannot effectively combine vehicle status, operating conditions, and identity characteristics for unified modeling, resulting in insufficient personalization of instrument displays and alarm controls, as well as inadequate security.
By employing multimodal biometric modeling, decomposing reward structures, and using an improved MaxEnt IRL policy learning method, a base trajectory and differential features are constructed, a basic reward and personalized reward decomposition mechanism is introduced, and a stable instrument control strategy is generated using time sparsity constraints. This enables adaptive interface updates and alarm outputs based on driver characteristics and operating conditions after successful identity verification.
It improves the robustness of vehicle identity authentication, reduces the risk of unauthorized startup, ensures vehicle access security, and reduces frequent interface changes and alarm triggers through personalized control strategies, thereby reducing the cognitive burden on drivers.
Smart Images
Figure CN121716643B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent industrial vehicle control technology, and in particular to an industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison. Background Technology
[0002] Industrial vehicles are widely used in warehousing, ports, manufacturing and other scenarios. Due to safety and management requirements, existing technologies generally use mechanical keys, IC cards, keypads or single biometrics for identification and unlocking control. Some systems have introduced face verification based on image recognition or single-modal identity authentication based on fingerprint comparison to restrict unauthorized personnel from starting the vehicle. These solutions generally complete a one-time identity verification before the vehicle is ignited. After successful verification, the power supply or control circuit is directly unlocked. The identity information is not continuously used during subsequent vehicle operation. The identity recognition results are rarely systematically correlated with the vehicle's operating status, instrument display logic and alarm behavior. They mostly remain at the access control level of "whether it can be started".
[0003] In terms of vehicle operation monitoring and instrument control, existing industrial vehicles mostly adopt fixed rules or simple threshold logic, triggering alarms and switching interface displays based on parameters such as vehicle speed, load, power, and hydraulic pressure. Some systems use historical data to optimize alarm thresholds offline or introduce simple statistical models to adjust alarm frequency, but the overall strategy is still uniform. It lacks characterization of differences in the operating styles, risk preferences, and attention allocation of different drivers. In existing technologies, even if trajectory-based analysis exists, it is mostly used for offline safety assessment or driving behavior scoring. It does not model vehicle status, operating conditions, instrument control actions, and identity characteristics together as a trajectory set that can be used for online decision optimization. Furthermore, it does not introduce baseline trajectories and differential features to provide targeted quantitative representation of the deviation between driving behavior and standard control logic.
[0004] At the level of intelligent decision-making and strategy learning, existing methods mostly focus on path planning, energy consumption optimization, or vehicle control, paying less attention to the decision-making process of instrument display and alarm control. They generally lack a decomposition modeling mechanism that breaks down the reward function into basic rewards and personalized rewards, making it difficult to distinguish the different contributions of operating condition factors and identity factors to the control strategy. At the same time, instrument control strategies often do not specifically constrain temporal sparsity, and alarms and interface updates are prone to dense triggering under high-frequency changing operating conditions, increasing the cognitive burden on the driver. Existing identity recognition and unlocking systems usually do not have the ability to combine vehicle running trajectory, differential features, and inverse strategy learning algorithms for strategy optimization, nor do they generate instrument display commands and alarm commands that take into account security, personalization, and temporal sparsity characteristics based on a unified framework after identity verification.
[0005] Therefore, how to provide an industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] One objective of this invention is to propose an industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison. This invention combines multimodal biometric modeling, reward structure decomposition, and an improved MaxEnt IRL policy learning method to achieve integrated processing of vehicle identity verification, unlocking control, and intelligent instrument display decision-making. By constructing baseline trajectories and differential features, introducing a basic reward and personalized reward decomposition mechanism, and using time sparse constraints to generate stable instrument control strategies, the system enables the vehicle to adaptively update the interface and output alarms based on driver characteristics and operating conditions after identity verification. This results in higher security, lower driving interference, and more personalized control strategies.
[0007] An industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to an embodiment of the present invention includes:
[0008] The multimodal biometric modeling module is used to collect facial images, voiceprints, and fingerprints and jointly model them to generate potential identity vectors.
[0009] The identity verification and unlocking control module is used to perform identity recognition and authorization comparison based on facial images, fingerprints and identity potential vectors, generate identity verification results, and output vehicle unlocking control commands to complete the unlocking operation;
[0010] The operation trajectory construction module is used to collect vehicle status data, operating condition data and instrument control action data, align them on a unified time axis and combine them with identity potential vectors to construct an operation trajectory set;
[0011] The baseline trajectory generation module is used to generate a set of baseline trajectories based on the vehicle dynamics model and fixed instrument control rules, and to calculate the difference feature sequence between the running trajectory set and the baseline trajectory set.
[0012] The decomposed reward structure construction module is used to construct a decomposed reward structure based on the set of running trajectories and differential feature sequences, and generate basic rewards and personalized rewards.
[0013] The MaxEnt IRL learning module is used to build an improved MaxEntIRL model based on differential feature sequences and decomposed reward structures, and to learn basic reward parameters and personalized reward parameters.
[0014] The time-sparse strategy solution module is used to introduce time-sparse constraints into the MaxEnt IRL model and solve the time-sparse instrument control strategy based on the event triggering conditions and the constant time interval conditions.
[0015] The instrument control execution module is used to generate instrument display commands and alarm commands based on time-sparse instrument control strategies, current vehicle status data, operating condition data, and identity potential vectors after authentication and unlocking, and to drive the instrument interface update and alarm output.
[0016] Optionally, modules can be integrated using the following methods:
[0017] Collect facial images, voiceprints, and fingerprints during driving, jointly model multimodal biometrics, and generate potential identity vectors;
[0018] Based on facial images, fingerprints, and potential identity vectors, the system performs identity recognition and authorization comparison, generates identity verification results, and outputs vehicle unlocking control commands to complete the unlocking operation.
[0019] A set of operating trajectories is constructed based on vehicle status, operating conditions, instrument control actions, and potential identity vectors.
[0020] A baseline trajectory set is generated based on the vehicle dynamics model and fixed instrument control rules, and the difference feature sequence between the running trajectory set and the baseline trajectory set is calculated.
[0021] Based on the set of operating trajectories and differential feature sequences, a decomposed reward structure is constructed, which splits the reward function into basic reward and personalized reward. The basic reward depends on vehicle status, operating conditions and instrument control actions, while the personalized reward depends on the identity latent vector.
[0022] An improved MaxEnt IRL model is built based on differential feature sequences and decomposed reward structure to learn basic reward parameters and personalized reward parameters.
[0023] In the MaxEnt IRL model, time sparsity constraints are introduced to set event triggering conditions and time interval constant conditions for the instrument control action sequence. The strength of time sparsity constraints is adjusted according to the operating conditions to solve the time sparsity instrument control strategy.
[0024] During vehicle operation after authentication and unlocking, instrument display commands and alarm commands are generated based on time-sparse instrument control strategy, vehicle status, operating conditions and identity potential vector, driving instrument interface updates and alarm output.
[0025] Optionally, the generation of the authentication result specifically includes:
[0026] The collected face images are preprocessed and face feature vectors are extracted; the collected fingerprints are preprocessed and fingerprint feature vectors are extracted.
[0027] Based on the mapping relationship between facial feature vectors, fingerprint feature vectors, and potential identity vectors, facial similarity scores and fingerprint similarity scores are calculated.
[0028] Read the target face template, target fingerprint template, and authorized identity identifier from the authorization information, and calculate the consistency score between the latent identity vector and the authorized identity identifier;
[0029] A comprehensive identity score is obtained by weighting and fusing the facial similarity score, fingerprint similarity score, and consistency score.
[0030] The overall identity score is compared with a preset verification threshold. If the overall identity score is not less than the preset verification threshold, a successful identity verification result is generated. If the overall identity score is less than the preset verification threshold, a failed identity verification result is generated.
[0031] Optionally, the construction of the set of running trajectories specifically includes:
[0032] Vehicle status data is collected from the vehicle control unit and on-board sensors and arranged in chronological order to form a vehicle status data sequence.
[0033] Operational condition data is collected from the operation control unit and environmental sensing device and arranged in chronological order to form an operational condition data sequence.
[0034] Data on instrument control actions are collected from the instrument system of industrial vehicles and arranged in chronological order to form a sequence of instrument control action data.
[0035] Time alignment is performed on the vehicle status data sequence, operating condition data sequence, and instrument control action data sequence on a unified time axis, and the vehicle status data, operating condition data, and instrument control action data at the same sampling time are combined to form an operating trajectory time slice;
[0036] Each operation trajectory is configured with an identity latent vector, which is a vector generated by joint modeling of multimodal biometrics. It is used to represent the identity features of the drivers participating in the operation trajectory and remains unchanged within the operation trajectory.
[0037] The running trajectories are arranged in chronological order to form time slices, and the running trajectories are combined into a set of running trajectories.
[0038] Optionally, the calculation of the differential feature sequence specifically includes:
[0039] A vehicle dynamics model is established, which takes vehicle state data and operating condition data in the time slice of the running trajectory as input and outputs the baseline vehicle state data of the corresponding time step.
[0040] Establish fixed instrument control rules. The fixed instrument control rules take the vehicle status data and operating condition data in the time slice of the running trajectory as input and output the baseline instrument control action data of the corresponding time step.
[0041] For each running trajectory in the running trajectory set, the vehicle status data and operating condition data in the time slice of the running trajectory are read in chronological order. The vehicle status data and operating condition data are input into the vehicle dynamics model to obtain baseline vehicle status data. The vehicle status data and operating condition data are input into the fixed instrument control rules to obtain baseline instrument control action data. The baseline vehicle status data, operating condition data and baseline instrument control action data are arranged in chronological order to form a baseline trajectory. All baseline trajectories constitute a baseline trajectory set.
[0042] At each time step, the difference between the vehicle status data in the running trajectory and the baseline vehicle status data in the baseline trajectory is calculated. At the same time step, the difference between the instrument control action data in the running trajectory and the baseline instrument control action data in the baseline trajectory is calculated. The vehicle status data difference and the instrument control action data difference are combined to form a differential feature. The differential features are arranged in time order to form a differential feature sequence, and a differential feature sequence is generated for each running trajectory in the running trajectory set.
[0043] Optionally, the construction of the decomposed reward structure and the splitting of the reward function specifically include:
[0044] Vehicle status data, operating condition data, instrument control action data, and identity potential vectors are read from the time slices of the operating trajectory set in chronological order. Differential features corresponding to the time slices of the operating trajectory are read from the differential feature sequence. The vehicle status data, operating condition data, instrument control action data, identity potential vectors, and differential features are combined to form the reward modeling input data.
[0045] Define a reward function that takes reward modeling input data as input and time step reward value as output. Inside the reward function, set a basic reward component and a personalized reward component. The basic reward component takes vehicle status data, operating condition data, instrument control action data and differential features as input, and the personalized reward component takes the identity potential vector as input.
[0046] Set basic reward parameters for the basic reward component and personalized reward parameters for the personalized reward component. Set decoupling constraints in the parameter space so that the basic reward component depends on the basic reward parameters, vehicle status data, operating condition data, instrument control action data and differential features, and the personalized reward component depends on the personalized reward parameters and the identity potential vector.
[0047] At each time step, the basic return component and the personalized return component are calculated. The basic return component and the personalized return component are added together to obtain the time step return value. The basic return components are arranged in time order to form a basic return sequence, and the personalized return components are arranged in time order to form a personalized return sequence. The basic return sequence and the personalized return sequence constitute the decomposed reward structure.
[0048] Optionally, the establishment and processing of the improved MaxEnt IRL model specifically includes:
[0049] Define the state space, action space, and trajectory space. The state consists of vehicle state data and operating condition data, the action consists of instrument control action data, and the trajectory consists of the running trajectories in the running trajectory set.
[0050] Based on the set of running trajectories and the corresponding differential feature sequences, the basic reward component and personalized reward component of each time step are calculated in chronological order. The basic reward component and personalized reward component are added together to obtain the time step reward value. The time step reward values on each running trajectory are accumulated to form the trajectory reward.
[0051] An improved MaxEnt IRL model is established. The MaxEnt IRL model takes the state space, action space and trajectory space as the modeling objects, takes the trajectory reward as the trajectory score, constructs the trajectory probability distribution in an exponential form based on the trajectory score, and normalizes the sum of all trajectory probabilities to one.
[0052] Construct a log-likelihood objective function, which takes the trajectory probabilities of the set of running trajectories as input and the sum of the log probabilities of the set of running trajectories under the improved MaxEnt IRL model as the objective value.
[0053] Based on the gradient information of the log-likelihood objective function with respect to the basic reward parameters and personalized reward parameters, the basic reward parameters and personalized reward parameters are iteratively updated until the log-likelihood objective function satisfies the convergence condition. The basic reward parameters and personalized reward parameters obtained at convergence are then used as the basic reward parameters and personalized reward parameters in the improved MaxEnt IRL model.
[0054] Optionally, the generation of the time-sparse instrument control strategy specifically includes:
[0055] In the improved MaxEnt IRL model, a corresponding instrument control action sequence variable is set for each running trajectory in the running trajectory set, and the instrument control action sequence variable is arranged according to time steps;
[0056] The event trigger index is calculated based on the vehicle status data and operating condition data in the time slice of the running trajectory. The event trigger time point is set at the time step when the event trigger index reaches the preset event trigger threshold. The time interval is divided on the time axis with adjacent event trigger time points as boundaries.
[0057] Apply a time interval constant condition within each time interval to keep the instrument control action sequence variables of all time steps within the time interval at the same value, and allow the instrument control action sequence variables to change values at the boundary of adjacent time intervals;
[0058] The number of times the instrument control action sequence variable changes between adjacent time intervals is counted on the time axis. The number of changes is multiplied by the time sparsity penalty coefficient to obtain the time sparsity penalty value, which represents the strength of the time sparsity constraint.
[0059] The operating condition load index is calculated based on the operating condition data. When the operating condition load index is higher than the first load threshold, the time sparsity penalty coefficient is increased. When the operating condition load index is lower than the second load threshold, the time sparsity penalty coefficient is decreased. When the operating condition load index is between the first load threshold and the second load threshold, the time sparsity penalty coefficient remains unchanged.
[0060] The time sparsity penalty value is added to the log-likelihood objective function of the improved MaxEnt IRL model to form a joint objective function containing the log-likelihood term and the time sparsity term. An iterative optimization algorithm is used to update the instrument control action sequence variables under the premise of satisfying the event triggering condition and the constant time interval condition. When the joint objective function satisfies the convergence criterion, the corresponding instrument control action sequence variable is used as the time sparsity instrument control strategy.
[0061] Optionally, the updating of the instrument interface and alarm output specifically includes:
[0062] After the vehicle is ignited, the authentication result is received. When the authentication result is successful, the vehicle unlock status signal is received and the time-sparse instrument control strategy is activated.
[0063] During vehicle operation, vehicle status data, operating condition data, and potential identity vectors are read according to the sampling period, and the vehicle status data, operating condition data, and potential identity vectors are combined in a preset order to form strategy input data.
[0064] Input the strategy input data into the time-sparse instrument control strategy, output the instrument control action of the current sampling period, and break down the instrument control action into instrument display instructions and alarm instructions;
[0065] Send instrument display commands to the instrument system to control the instrument interface page layout, information display content, brightness level, and indicator icon display method;
[0066] The alarm command is sent to the alarm execution module to control the start / stop, alarm level and duration of the audible and visual alarm devices.
[0067] When the authentication result becomes unsuccessful, the vehicle unlock status is released, or the ignition signal disappears, the output of the time-sparse instrument control strategy is stopped, and the instrument interface and alarm output are restored to the safe shutdown display state.
[0068] The beneficial effects of this invention are:
[0069] This invention, through the collaboration of a multimodal biometric modeling module and an identity verification and unlocking control module, enables facial images, voiceprints, and fingerprints to be used not only for one-time vehicle start permission judgment, but also to establish a unified representation with authorization information in the form of identity latent vectors. This achieves consistency assessment and comprehensive identity score calculation based on facial features, fingerprint features, and identity latent vectors. Compared with existing solutions that rely solely on single image recognition or fingerprint comparison, this invention can simultaneously utilize multi-source features and consistency constraints in the latent vector space during the identity verification process, improving the robustness of identity authentication and reducing the risk of illegal vehicle start-up such as card theft and forgery of single features. Unlocking control commands are only output after identity verification is successful, significantly enhancing vehicle access security.
[0070] This invention encapsulates vehicle status, operating conditions, instrument control actions, and identity latent vectors into a unified operating trajectory through an operating trajectory construction module and a baseline trajectory generation module. This trajectory is then combined with the differential feature sequence between the baseline trajectory and the operating trajectory. In the reward structure decomposition module, the reward function is split into basic reward and personalized reward. The MaxEntIRL learning module then learns the basic reward parameters and personalized reward parameters, enabling the system to distinguish the different impacts of operating condition driving factors and driver identity preferences on instrument control behavior. Based on this, the time sparse strategy solution module applies time sparse constraints to instrument control actions through event triggering conditions and constant time interval conditions. This suppresses frequent and low-value interface changes and alarm triggers while ensuring timely alarms for critical operating conditions. As a result, the instrument display instructions and alarm instructions output by the instrument control execution module after identity verification and unlocking are more focused on high-value information. Attached Figure Description
[0071] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0072] Figure 1 This is a structural diagram of the industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison proposed in this invention;
[0073] Figure 2 This is a flowchart of the industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison proposed in this invention;
[0074] Figure 3 This is a schematic diagram illustrating the generation of a time-sparse instrument control strategy for an industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison, as proposed in this invention. Detailed Implementation
[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0076] refer to Figure 1-3 An industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison includes:
[0077] The multimodal biometric modeling module is used to collect facial images, voiceprints, and fingerprints and jointly model them to generate potential identity vectors.
[0078] The identity verification and unlocking control module is used to perform identity recognition and authorization comparison based on facial images, fingerprints and identity potential vectors, generate identity verification results, and output vehicle unlocking control commands to complete the unlocking operation;
[0079] The operation trajectory construction module is used to collect vehicle status data, operating condition data and instrument control action data, align them on a unified time axis and combine them with identity potential vectors to construct an operation trajectory set;
[0080] The baseline trajectory generation module is used to generate a set of baseline trajectories based on the vehicle dynamics model and fixed instrument control rules, and to calculate the difference feature sequence between the running trajectory set and the baseline trajectory set.
[0081] The decomposed reward structure construction module is used to construct a decomposed reward structure based on the set of running trajectories and differential feature sequences, and generate basic rewards and personalized rewards.
[0082] The MaxEnt IRL learning module is used to build an improved MaxEntIRL model based on differential feature sequences and decomposed reward structures, and to learn basic reward parameters and personalized reward parameters.
[0083] The time-sparse strategy solution module is used to introduce time-sparse constraints into the MaxEnt IRL model and solve the time-sparse instrument control strategy based on the event triggering conditions and the constant time interval conditions.
[0084] The instrument control execution module is used to generate instrument display commands and alarm commands based on time-sparse instrument control strategies, current vehicle status data, operating condition data, and identity potential vectors after authentication and unlocking, and to drive the instrument interface update and alarm output.
[0085] In this embodiment, the modules are interconnected using the following method:
[0086] Collect facial images, voiceprints, and fingerprints during driving, jointly model multimodal biometrics, and generate potential identity vectors;
[0087] A set of operating trajectories is constructed based on vehicle status, operating conditions, instrument control actions, and potential identity vectors.
[0088] A baseline trajectory set is generated based on the vehicle dynamics model and fixed instrument control rules, and the difference feature sequence between the running trajectory set and the baseline trajectory set is calculated.
[0089] Based on the set of operating trajectories and differential feature sequences, a decomposed reward structure is constructed, which splits the reward function into basic reward and personalized reward. The basic reward depends on vehicle status, operating conditions and instrument control actions, while the personalized reward depends on the identity latent vector.
[0090] An improved MaxEnt IRL model is built based on differential feature sequences and decomposed reward structure to learn basic reward parameters and personalized reward parameters.
[0091] In the MaxEnt IRL model, time sparsity constraints are introduced to set event triggering conditions and time interval constant conditions for the instrument control action sequence. The strength of time sparsity constraints is adjusted according to the operating conditions to solve the time sparsity instrument control strategy.
[0092] During the operation of industrial vehicles, instrument display control commands and alarm control commands are generated based on the time-sparse instrument control strategy, the current vehicle status, operating conditions, and potential identity vectors, and the instrument interface and alarm output are updated.
[0093] In this embodiment, the generation of the authentication result specifically includes:
[0094] The acquired face images are subjected to size normalization, illumination compensation and noise suppression processing. The processed face images are then input into the face feature extraction network, which outputs a face feature vector. Each component in the face feature vector is used to characterize the numerical description of the face in different feature dimensions.
[0095] The acquired fingerprint image is processed by grayscale enhancement, ridge thinning and artifact removal. The processed fingerprint image is then input into the fingerprint feature extraction network, which outputs a fingerprint feature vector. Each component in the fingerprint feature vector is used to characterize the numerical description of the fingerprint texture in different feature dimensions.
[0096] Based on the pre-trained mapping relationship between face feature vector, fingerprint feature vector and potential identity vector, the distance metric between the face feature vector and the target face template feature vector is calculated in the feature space, and the distance metric is mapped to the face similarity score. Similarly, the distance metric between the fingerprint feature vector and the target fingerprint template feature vector is calculated in the feature space, and the distance metric is mapped to the fingerprint similarity score. The face similarity score and fingerprint similarity score are used to quantify the proximity of the current face and fingerprint to the authorized object.
[0097] The face template, fingerprint template, and authorized identity identifier bound to the industrial vehicle are read from the authorization information. The authorized identity identifier is represented as the authorized identity potential vector through the same identity potential vector modeling process. The consistency measure between the current identity potential vector and the authorized identity potential vector is calculated in the identity potential vector space. The consistency measure is mapped to a consistency score. The consistency score is used to quantify the degree of consistency between the current multimodal joint identity features and the authorized identity.
[0098] Normalization is performed on the face similarity score, fingerprint similarity score, and consistency score. Weight coefficients are set for the face similarity score, fingerprint similarity score, and consistency score respectively. The normalized score is multiplied by the corresponding weight coefficient and summed to obtain the comprehensive identity score. The comprehensive identity score is used to give an overall credibility evaluation of the current driver's identity matching the authorized identity.
[0099] The overall identity score is compared with a preset verification threshold. When the overall identity score is greater than or equal to the preset verification threshold, a successful identity verification result is generated, and a pass mark and the corresponding authorized identity identifier are recorded in the identity verification result. When the overall identity score is less than the preset verification threshold, a failed identity verification result is generated, and a fail mark and a rejection reason mark are recorded in the identity verification result. The identity verification result is used to control the output of vehicle unlocking control commands.
[0100] In this embodiment, the construction of the set of running trajectories specifically includes:
[0101] A vehicle control unit is set up in the vehicle electronic control system. The vehicle control unit is connected to the vehicle speed sensor, acceleration sensor, steering angle sensor, brake signal acquisition circuit, drive system controller, and gear position sensor. The vehicle operating parameters are collected within a preset sampling period to form vehicle status data. The vehicle status data is timestamped and preprocessed, and then arranged in chronological order to form a vehicle status data sequence.
[0102] An operation control unit is set up in the operation control system. The operation control unit is connected to the load sensor, operation mode selection device, positioning device, road condition acquisition device, and ambient light sensor. Within a preset sampling period, it collects operating condition data such as load information, operation mode information, operation position, road slope, ground adhesion, and ambient brightness. The operating condition data is timestamped and preprocessed, and arranged in chronological order to form an operating condition data sequence.
[0103] Data on instrument control actions are collected from the instrument system of industrial vehicles. Within a preset sampling period, instrument control actions such as page switching, alarm level setting, display brightness adjustment, and indicator icon switching are recorded. Timestamps are added to the instrument control action data and a unified format is performed. The data is then arranged in chronological order to form a sequence of instrument control action data.
[0104] Time alignment is performed on the vehicle status data sequence, operating condition data sequence and instrument control action data sequence on a unified time axis. A unified sampling period is used as the time step. The vehicle status data, operating condition data and instrument control action data corresponding to the same time step are combined to form the running trajectory time slice.
[0105] Each operation trajectory is configured with an identity potential vector, which is a vector generated by multimodal biometric joint modeling based on face image, voiceprint signal and fingerprint image. It is used to represent the identity features of the driver participating in the operation trajectory and remains unchanged in all operation trajectory time slices contained in the operation trajectory.
[0106] The running trajectories are arranged in chronological order to form time slices, and the running trajectories are then collected to form a running trajectory set.
[0107] In this embodiment, the calculation of the differential feature sequence specifically includes:
[0108] A vehicle dynamics model is established by selecting vehicle state data and operating condition data from the set of operating trajectories as modeling samples. The continuous time vehicle state change process is discretized into a time step sequence according to the preset sampling period. In each time step, the vehicle state data and operating condition data form an input vector, and the vehicle state data of the next time step is used as the output vector. The system identification algorithm is used to train a vehicle dynamics model with vehicle state data and operating condition data as input and baseline vehicle state data as output. State boundaries and operating condition boundaries are set in the model to limit the prediction results.
[0109] Establish fixed instrument control rules. Based on safety regulations and experience-based control logic, divide vehicle status data and operating condition data into several control areas. Preset corresponding instrument control actions in each control area and establish a mapping relationship between control areas and control actions to form fixed instrument control rules with vehicle status data and operating condition data as input and output baseline instrument control action data.
[0110] For each running trajectory in the running trajectory set, the vehicle status data and operating condition data in the time slice of the running trajectory are read in chronological order. In each time step, the vehicle status data and operating condition data are input into the vehicle dynamics model to obtain baseline vehicle status data. The vehicle status data and operating condition data of the same time step are input into the fixed instrument control rules to obtain baseline instrument control action data. The obtained baseline vehicle status data and baseline instrument control action data are arranged in chronological order to form a baseline trajectory. All baseline trajectories are combined into a baseline trajectory set.
[0111] Within each time step, the vehicle status data in the running trajectory is subtracted from the baseline vehicle status data in the baseline trajectory to obtain vehicle status differential data. The instrument control action data in the running trajectory is subtracted from the baseline instrument control action data in the baseline trajectory to obtain instrument control action differential data. The vehicle status differential data and the instrument control action differential data are concatenated in a preset order to form differential features. The differential features are arranged sequentially on the time axis to form a differential feature sequence, generating a differential feature sequence for each running trajectory in the running trajectory set.
[0112] In this embodiment, the construction of the decomposed reward structure and the splitting of the reward function specifically include:
[0113] Vehicle status data, operating condition data, instrument control action data, and identity potential vectors are read from the time slices of the operating trajectory set in chronological order. Differential features corresponding to the time slices of the operating trajectory are read from the differential feature sequence. Dimension alignment and numerical normalization are performed on the vehicle status data, operating condition data, instrument control action data, and differential features. The processed vehicle status data, operating condition data, instrument control action data, identity potential vectors, and differential features are combined to form the input data for reward modeling.
[0114] Define a reward function that takes reward modeling input data as input and time step reward value as output. Set a basic reward component and a personalized reward component inside the reward function. The basic reward component takes vehicle status data, operating condition data, instrument control action data and differential features as input, and the personalized reward component takes the identity latent vector as input.
[0115] A set of basic reward parameters is set for the basic reward components. The set of basic reward parameters includes basic reward weight parameters and basic reward bias parameters. In each time step, vehicle status data, operating condition data, instrument control action data and differential features are expanded into basic reward input vectors in a preset order. Each component in the basic reward input vector is multiplied by the corresponding basic reward weight parameter and summed. The basic reward bias parameter is then added, and the result is converted into a scalar value of the basic reward component through a monotonic nonlinear mapping function.
[0116] A set of personalized reward parameters is set for the personalized reward components. The set of personalized reward parameters includes personalized reward weight parameters and personalized reward bias parameters. At each time step, the identity potential vector is used as the personalized reward input vector. Each component in the personalized reward input vector is multiplied by the corresponding personalized reward weight parameter and summed. The personalized reward bias parameter is then added and converted into a scalar value of the personalized reward component through a monotonic nonlinear mapping function.
[0117] Decoupling constraints are set between the basic reward parameter set and the personalized reward parameter set. By managing the basic reward parameter set and the personalized reward parameter set in the storage structure and parameter update rules respectively, the calculation process of the basic reward component depends only on the basic reward parameter set, vehicle status data, operating condition data, instrument control action data and differential features, and the calculation process of the personalized reward component depends only on the personalized reward parameter set and the identity potential vector.
[0118] Within each time step, the scalar value of the basic reward component is added to the scalar value of the personalized reward component to obtain the time step reward value. The scalar values of the basic reward component are arranged in chronological order to form the basic reward sequence, and the scalar values of the personalized reward component are arranged in chronological order to form the personalized reward sequence. The basic reward sequence and the personalized reward sequence constitute the decomposed reward structure.
[0119] In this embodiment, the establishment and processing of the improved MaxEnt IRL model specifically includes:
[0120] The state space is defined based on the vehicle status data and operating condition data of the time slices of the running trajectory in the running trajectory set; the action space is defined based on the instrument control action data of the time slices of the running trajectory; and the trajectory space is defined based on the time sequence arrangement of the running trajectories in the running trajectory set.
[0121] For each running trajectory in the running trajectory set, the time step reward value is read in chronological order. The time step reward value is obtained by adding the basic reward component and the personalized reward component. The trajectory reward is obtained by adding all the time step reward values on the running trajectory. All trajectory rewards are associated with the corresponding running trajectories to form a trajectory reward set.
[0122] An improved MaxEnt IRL model is established. The improved MaxEnt IRL model takes the state space, action space and trajectory space as modeling objects, and takes the trajectory reward in the trajectory reward set as the trajectory score. Based on the trajectory score, the unnormalized probability value of each running trajectory is calculated through exponential mapping. The normalization factor is obtained by summing all the unnormalized probability values, and the trajectory probability distribution is obtained by dividing the unnormalized probability value by the normalization factor.
[0123] Construct a log-likelihood objective function. The log-likelihood objective function takes the running trajectories and their probability distributions in the set of running trajectories as inputs, calculates the sum of the log probabilities of all running trajectories under the improved MaxEnt IRL model, and uses this sum as the value of the log-likelihood objective function.
[0124] The gradient information of the log-likelihood objective function with respect to the base reward parameters and personalized reward parameters is calculated. Based on the gradient information, the base reward parameters and personalized reward parameters are adjusted using an iterative update method. In each iteration, the time step reward value, trajectory reward, trajectory probability distribution, and log-likelihood objective function value are recalculated using the current base reward parameters and personalized reward parameters. The iteration stops when the change in the log-likelihood objective function value is less than the convergence threshold. The base reward parameters and personalized reward parameters obtained at this time are used as the base reward parameters and personalized reward parameters of the improved MaxEnt IRL model.
[0125] In this embodiment, the generation of the time-sparse instrument control strategy specifically includes:
[0126] In the improved MaxEnt IRL model, instrument control action sequence variables are set for each running trajectory in the running trajectory set. The instrument control action sequence variables are arranged in time steps on the time axis, and each time step corresponds to an instrument control action value to be optimized.
[0127] The event trigger index is calculated based on the vehicle status data and operating condition data in the time slice of the running trajectory. The event trigger index is obtained by superimposing the changes in vehicle speed, steering angle, brake pedal travel, load, and ambient brightness according to preset weights. The event trigger time point is set at the time step where the event trigger index is greater than the event trigger threshold, and the time interval is divided on the time axis with adjacent event trigger time points as boundaries.
[0128] Apply a time interval constant condition within each time interval to constrain the instrument control action sequence variables of all time steps within that time interval to the same value. Allow the values of the instrument control action sequence variables to change at the boundary of adjacent time intervals, thereby forming a segmented constant instrument control action sequence structure on the time axis.
[0129] The number of times the values of the instrument control action sequence variables change between adjacent time intervals is counted on the time axis. The time sparsity penalty value is obtained by multiplying the number of value changes by the time sparsity penalty coefficient. The time sparsity penalty value is used to measure the frequency of change of the instrument control action over time.
[0130] The operating condition load index is calculated based on the operating condition data. The operating condition load index is obtained by combining the load level, operation mode, road slope and braking operation frequency according to the preset weight. When the operating condition load index is higher than the first load threshold, the time sparsity penalty coefficient is increased. When the operating condition load index is lower than the second load threshold, the time sparsity penalty coefficient is decreased. When the operating condition load index is between the first load threshold and the second load threshold, the time sparsity penalty coefficient remains unchanged.
[0131] The time sparsity penalty is introduced into the log-likelihood objective function of the improved MaxEnt IRL model, forming a joint objective function containing the log-likelihood term and the time sparsity penalty term. In each iteration, the time step reward value, trajectory reward, and trajectory probability distribution are calculated based on the current instrument control action sequence variable, and then the joint objective function value is calculated. The instrument control action sequence variable is updated according to the gradient information of the joint objective function value with respect to the instrument control action sequence variable. The iteration stops when the change of the joint objective function value between two consecutive iterations is lower than the convergence threshold. The instrument control action sequence variable obtained at this time is used as the time sparsity instrument control strategy.
[0132] In this embodiment, updating the instrument interface and alarm output specifically includes:
[0133] When the vehicle is ignited, the authentication result flag and the vehicle unlock status flag are received. When the authentication result flag is equal to the pass flag and the vehicle unlock status flag is equal to the unlock flag, the enable flag of the time sparse instrument control strategy is set to be valid. When the enable flag is invalid, the instrument interface is kept in a safe standby display state and alarm output is disabled.
[0134] During the effective period of the activation mark, vehicle status data is read from the vehicle control unit according to the sampling cycle, operating condition data is read from the operation control unit and the environmental perception device, and identity potential vector is read from the multimodal biometric module. The vehicle status data and operating condition data are standardized, and the standardized vehicle status data, operating condition data and identity potential vector are spliced together in a preset order to form a strategy input vector.
[0135] The strategy input vector is input into the time-sparse instrument control strategy. Within the time-sparse instrument control strategy, the time interval to which it belongs is determined based on the time index of the current time step. The constant instrument control action of the time interval is read, and the constant instrument control action of the time interval is mapped to the instrument control action of the current time step. The instrument control action is split into instrument display command component and alarm command component.
[0136] The instrument display command component is sent to the instrument system. Inside the instrument system, the interface page template is selected according to the instrument display command component, the page layout, numerical display area, icon display area and brightness level are updated, and the updated interface buffer content is written to the display driver and output to the instrument screen.
[0137] The alarm command component is sent to the alarm execution module. Inside the alarm execution module, the sound alarm mode and light alarm mode are selected according to the alarm command component, the alarm level, flashing frequency and duration are set, and the buzzer and warning light are controlled to execute the alarm output.
[0138] During the sampling period, the system continuously monitors the authentication result flag, vehicle unlock status flag, and ignition signal. When the authentication result flag is not equal to the pass flag, or the vehicle unlock status flag is not equal to the unlock flag, or the ignition signal is in the off state, the enable flag is set to invalid, and an instrument reset command and an alarm stop command are sent to make the instrument interface enter the safe stop display state and stop the alarm output.
[0139] Example 1:
[0140] To verify the feasibility of this invention in practice, it was applied to a warehouse logistics operation scenario using electric forklifts and tractor-trailers. The existing vehicle management method was as follows: drivers used mechanical keys to turn on the power and then completed authorization verification by swiping an IC card. Some main lane vehicles were additionally equipped with single facial recognition terminals for facial verification before ignition. Identity information was only used during vehicle startup; once verified, it was no longer associated with the operation process. Any cardholder or person with a high degree of facial similarity to the template could drive continuously for extended periods. The instrument control system used fixed threshold logic and a unified interface layout, setting single alarm thresholds for parameters such as vehicle speed, load, and battery level. Exceeding these thresholds triggered audible and visual alarms and flashing icons, and all drivers saw the same page structure. After long-term operation, it was found that instances of drivers using borrowed cards to start vehicles occurred frequently. Some drivers, especially when heavily loaded and turning, or when slightly speeding for extended periods, generated excessively high alarm frequencies, creating "alarm noise." Drivers habitually ignored these warnings. Although management could review logs afterward, it was difficult to distinguish the differences in risk behaviors among different drivers, and it was impossible to adjust the instrument display strategy according to individual habits.
[0141] Under the same workload and shift schedule, the industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison proposed in this invention was installed on some vehicles, forming a control group against the original key + IC card + fixed threshold alarm system. For vehicles using this invention, the driver stands in front of the vehicle terminal before getting in the car, and generates multimodal biometric input through face capture, fingerprint capture, and brief voice capture. The multimodal biometric modeling module jointly models the face image, voiceprint, and fingerprint to generate an identity potential vector. The identity verification and unlocking control module performs identity recognition and authorization comparison based on the face image, fingerprint, and identity potential vector. Only when the comprehensive score reaches the threshold is an unlocking control command output, allowing the vehicle to enter driving mode; otherwise, it remains locked. During vehicle operation, the trajectory construction module continuously records vehicle status, operating conditions, and instrument control actions, forming a trajectory set together with the identity latent vector. The baseline trajectory generation module generates a baseline trajectory set using the vehicle dynamics model and fixed instrument control rules, and calculates the differential feature sequence between the trajectory and the baseline trajectory. The reward structure decomposition module splits the reward function into basic reward and personalized reward, and the MaxEntIRL learning module learns the basic reward parameters and personalized reward parameters under the differential feature constraints. The time-sparse strategy solution module introduces time-sparse constraints into the improved MaxEnt IRL model, and solves the time-sparse instrument control strategy based on the event triggering conditions and the constant time interval conditions. During operation after identity verification and unlocking, the instrument control execution module generates instrument display instructions and alarm instructions based on the time-sparse instrument control strategy, vehicle status, operating conditions, and identity latent vector, and adjusts the page layout, information hierarchy, and audio-visual alarm output frequency.
[0142] Over several consecutive statistical periods, data was collected on vehicles using the original key + IC card + fixed threshold alarm system and vehicles using the system of this invention under conditions of similar mileage, similar cargo type, and similar shift duration. The key statistical indicators included the interception rate of unauthorized vehicle attempts, the proportion of legitimate drivers being falsely denied access, the number of alarm triggers per unit time, the proportion of alarms for non-critical operating conditions, the number of instrument panel page switches per shift, the proportion of alarms manually confirmed by the driver, the number of minor speeding and minor overloading events per shift, and the average loading and unloading volume per vehicle. Examples of the statistical results are shown in Table 1.
[0143] Table 1 Comparison of Industrial Vehicle Identification and Instrument Control Effects
[0144]
[0145] As shown in the table, the interception rate of unauthorized vehicle start attempts on vehicles using the system of this invention is significantly higher than that of the key + IC card + fixed threshold system. This indicates that the identity verification mechanism combining multimodal identity latent vectors with image recognition and fingerprint comparison is more effective in preventing unauthorized personnel from starting the vehicle. The proportion of false rejections by legitimate drivers has decreased, indicating that multimodal fusion has stronger robustness to interference such as changes in lighting and surface stains. The number of alarm triggers and the proportion of alarms in non-critical operating conditions have decreased significantly, the number of instrument panel page switching has decreased, while the proportion of alarms manually confirmed by the driver has increased. This indicates that the time-sparse instrument control strategy reduces alarm noise while ensuring key risk warnings, making drivers more sensitive to truly important alarms. The number of minor speeding and minor overloading events has decreased significantly, while the average loading and unloading workload per vehicle has slightly increased. This indicates that the strategy learned based on the decomposed reward structure and the improved MaxEnt IRL model guides drivers to form more robust operating behaviors without sacrificing efficiency. From a data perspective, this verifies the practical application value of this invention in the integration of identity security access and intelligent instrument control.
[0146] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison, characterized in that, include: The multimodal biometric modeling module is used to collect facial images, voiceprints, and fingerprints and jointly model them to generate potential identity vectors. The identity verification and unlocking control module is used to perform identity recognition and authorization comparison based on facial images, fingerprints and identity potential vectors, generate identity verification results, and output vehicle unlocking control commands to complete the unlocking operation; The operation trajectory construction module is used to collect vehicle status data, operating condition data and instrument control action data, align them on a unified time axis and combine them with identity potential vectors to construct an operation trajectory set; The baseline trajectory generation module is used to generate a set of baseline trajectories based on the vehicle dynamics model and fixed instrument control rules, and to calculate the difference feature sequence between the running trajectory set and the baseline trajectory set. The decomposed reward structure construction module is used to construct a decomposed reward structure based on the set of running trajectories and the differential feature sequence, and to generate basic rewards and personalized rewards, specifically: Vehicle status data, operating condition data, instrument control action data, and identity potential vectors are read from the time slices of the operating trajectory set in chronological order. Differential features corresponding to the time slices of the operating trajectory are read from the differential feature sequence. The vehicle status data, operating condition data, instrument control action data, identity potential vectors, and differential features are combined to form the reward modeling input data. Define a reward function that takes reward modeling input data as input and time step reward value as output. Inside the reward function, set a basic reward component and a personalized reward component. The basic reward component takes vehicle status data, operating condition data, instrument control action data and differential features as input, and the personalized reward component takes the identity potential vector as input. Set basic reward parameters for the basic reward component and personalized reward parameters for the personalized reward component. Set decoupling constraints in the parameter space so that the basic reward component depends on the basic reward parameters, vehicle status data, operating condition data, instrument control action data and differential features, and the personalized reward component depends on the personalized reward parameters and the identity potential vector. At each time step, the basic return component and the personalized return component are calculated. The basic return component and the personalized return component are added together to obtain the time step return value. The basic return components are arranged in time order to form a basic return sequence, and the personalized return components are arranged in time order to form a personalized return sequence. The basic return sequence and the personalized return sequence constitute the decomposed reward structure. The MaxEnt IRL learning module is used to build an improved MaxEnt IRL model based on differential feature sequences and decomposed reward structures, and to learn basic reward parameters and personalized reward parameters. The time-sparse strategy solution module is used to introduce time-sparse constraints into the MaxEnt IRL model and solve for time-sparse instrument control strategies based on event triggering conditions and constant time interval conditions. Specifically: In the improved MaxEnt IRL model, a corresponding instrument control action sequence variable is set for each running trajectory in the running trajectory set, and the instrument control action sequence variable is arranged according to time steps; The event trigger index is calculated based on the vehicle status data and operating condition data in the time slice of the running trajectory. The event trigger time point is set at the time step when the event trigger index reaches the preset event trigger threshold. The time interval is divided on the time axis with adjacent event trigger time points as boundaries. Apply a time interval constant condition within each time interval to keep the instrument control action sequence variables of all time steps within the time interval at the same value, and allow the instrument control action sequence variables to change values at the boundary of adjacent time intervals; The number of times the instrument control action sequence variable changes between adjacent time intervals is counted on the time axis. The number of changes is multiplied by the time sparsity penalty coefficient to obtain the time sparsity penalty value, which represents the strength of the time sparsity constraint. The operating condition load index is calculated based on the operating condition data. When the operating condition load index is higher than the first load threshold, the time sparsity penalty coefficient is increased. When the operating condition load index is lower than the second load threshold, the time sparsity penalty coefficient is decreased. When the operating condition load index is between the first load threshold and the second load threshold, the time sparsity penalty coefficient remains unchanged. The time sparsity penalty value is added to the log-likelihood objective function of the improved MaxEnt IRL model to form a joint objective function containing the log-likelihood term and the time sparsity term. An iterative optimization algorithm is used to update the instrument control action sequence variables under the premise of satisfying the event triggering condition and the constant time interval condition. When the joint objective function satisfies the convergence criterion, the corresponding instrument control action sequence variable is used as the time sparsity instrument control strategy. The instrument control execution module is used to generate instrument display commands and alarm commands based on time-sparse instrument control strategies, current vehicle status data, operating condition data, and identity potential vectors after authentication and unlocking, and to drive the instrument interface update and alarm output.
2. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 1, characterized in that, The modules are connected in the following way: Collect facial images, voiceprints, and fingerprints during driving, jointly model multimodal biometrics, and generate potential identity vectors; Based on facial images, fingerprints, and potential identity vectors, the system performs identity recognition and authorization comparison, generates identity verification results, and outputs vehicle unlocking control commands to complete the unlocking operation. A set of operating trajectories is constructed based on vehicle status, operating conditions, instrument control actions, and potential identity vectors. A baseline trajectory set is generated based on the vehicle dynamics model and fixed instrument control rules, and the difference feature sequence between the running trajectory set and the baseline trajectory set is calculated. Based on the set of operating trajectories and differential feature sequences, a decomposed reward structure is constructed, which splits the reward function into basic reward and personalized reward. The basic reward depends on vehicle status, operating conditions and instrument control actions, while the personalized reward depends on the identity latent vector. An improved MaxEnt IRL model is built based on differential feature sequences and decomposed reward structure to learn basic reward parameters and personalized reward parameters. In the MaxEnt IRL model, time sparsity constraints are introduced to set event triggering conditions and time interval constant conditions for the instrument control action sequence. The strength of time sparsity constraints is adjusted according to the operating conditions to solve the time sparsity instrument control strategy. During vehicle operation after authentication and unlocking, instrument display commands and alarm commands are generated based on time-sparse instrument control strategy, vehicle status, operating conditions and identity potential vector, driving instrument interface updates and alarm output.
3. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 2, characterized in that, The generation of the authentication result specifically includes: The collected face images are preprocessed and face feature vectors are extracted; the collected fingerprints are preprocessed and fingerprint feature vectors are extracted. Based on the mapping relationship between facial feature vectors, fingerprint feature vectors, and potential identity vectors, facial similarity scores and fingerprint similarity scores are calculated. Read the target face template, target fingerprint template, and authorized identity identifier from the authorization information, and calculate the consistency score between the latent identity vector and the authorized identity identifier; A comprehensive identity score is obtained by weighting and fusing the facial similarity score, fingerprint similarity score, and consistency score. The overall identity score is compared with a preset verification threshold. If the overall identity score is not less than the preset verification threshold, a successful identity verification result is generated. If the overall identity score is less than the preset verification threshold, a failed identity verification result is generated.
4. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 2, characterized in that, The construction of the set of running trajectories specifically includes: Vehicle status data is collected from the vehicle control unit and on-board sensors and arranged in chronological order to form a vehicle status data sequence. Operational condition data is collected from the operation control unit and environmental sensing device and arranged in chronological order to form an operational condition data sequence. Data on instrument control actions are collected from the instrument system of industrial vehicles and arranged in chronological order to form a sequence of instrument control action data. Time alignment is performed on the vehicle status data sequence, operating condition data sequence, and instrument control action data sequence on a unified time axis, and the vehicle status data, operating condition data, and instrument control action data at the same sampling time are combined to form an operating trajectory time slice; Each operation trajectory is configured with an identity latent vector, which is a vector generated by joint modeling of multimodal biometrics. It is used to represent the identity features of the drivers participating in the operation trajectory and remains unchanged within the operation trajectory. The running trajectories are arranged in chronological order to form time slices, and the running trajectories are combined into a set of running trajectories.
5. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 2, characterized in that, The calculation of the differential feature sequence specifically includes: A vehicle dynamics model is established, which takes vehicle state data and operating condition data in the time slice of the running trajectory as input and outputs the baseline vehicle state data of the corresponding time step. Establish fixed instrument control rules. The fixed instrument control rules take the vehicle status data and operating condition data in the time slice of the running trajectory as input and output the baseline instrument control action data of the corresponding time step. For each running trajectory in the running trajectory set, the vehicle status data and operating condition data in the time slice of the running trajectory are read in chronological order. The vehicle status data and operating condition data are input into the vehicle dynamics model to obtain baseline vehicle status data. The vehicle status data and operating condition data are input into the fixed instrument control rules to obtain baseline instrument control action data. The baseline vehicle status data, operating condition data and baseline instrument control action data are arranged in chronological order to form a baseline trajectory. All baseline trajectories constitute a baseline trajectory set. At each time step, the difference between the vehicle status data in the running trajectory and the baseline vehicle status data in the baseline trajectory is calculated. At the same time step, the difference between the instrument control action data in the running trajectory and the baseline instrument control action data in the baseline trajectory is calculated. The vehicle status data difference and the instrument control action data difference are combined to form a differential feature. The differential features are arranged in time order to form a differential feature sequence, and a differential feature sequence is generated for each running trajectory in the running trajectory set.
6. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 2, characterized in that, The establishment and processing of the improved MaxEnt IRL model specifically includes: Define the state space, action space, and trajectory space. The state consists of vehicle state data and operating condition data, the action consists of instrument control action data, and the trajectory consists of the running trajectories in the running trajectory set. Based on the set of running trajectories and the corresponding differential feature sequences, the basic reward component and personalized reward component of each time step are calculated in chronological order. The basic reward component and personalized reward component are added together to obtain the time step reward value. The time step reward values on each running trajectory are accumulated to form the trajectory reward. An improved MaxEnt IRL model is established. The MaxEnt IRL model takes the state space, action space and trajectory space as the modeling objects, takes the trajectory reward as the trajectory score, constructs the trajectory probability distribution in an exponential form based on the trajectory score, and normalizes the sum of all trajectory probabilities to one. Construct a log-likelihood objective function, which takes the trajectory probabilities of the set of running trajectories as input and the sum of the log probabilities of the set of running trajectories under the improved MaxEnt IRL model as the objective value. Based on the gradient information of the log-likelihood objective function with respect to the basic reward parameters and personalized reward parameters, the basic reward parameters and personalized reward parameters are iteratively updated until the log-likelihood objective function satisfies the convergence condition. The basic reward parameters and personalized reward parameters obtained at convergence are then used as the basic reward parameters and personalized reward parameters in the improved MaxEnt IRL model.
7. The industrial vehicle identification and unlocking system based on image recognition and fingerprint comparison according to claim 2, characterized in that, The updated instrument interface and alarm output specifically include: After the vehicle is ignited, the authentication result is received. When the authentication result is successful, the vehicle unlock status signal is received and the time-sparse instrument control strategy is activated. During vehicle operation, vehicle status data, operating condition data, and potential identity vectors are read according to the sampling period, and the vehicle status data, operating condition data, and potential identity vectors are combined in a preset order to form strategy input data. Input the strategy input data into the time-sparse instrument control strategy, output the instrument control action of the current sampling period, and break down the instrument control action into instrument display instructions and alarm instructions; Send instrument display commands to the instrument system to control the instrument interface page layout, information display content, brightness level, and indicator icon display method; The alarm command is sent to the alarm execution module to control the start / stop, alarm level and duration of the audible and visual alarm devices. When the authentication result becomes unsuccessful, the vehicle unlock status is released, or the ignition signal disappears, the output of the time-sparse instrument control strategy is stopped, and the instrument interface and alarm output are restored to the safe shutdown display state.