Terrain recognition method and system and electronic equipment

Abstract

The invention provides a terrain recognition method, which belongs to the field of intelligent driving and comprises the following steps: acquiring visual data, dynamic data and weather and vehicle working condition data; inputting the visual data into the recognition model to obtain a visual terrain scoring result, and inputting the dynamic data into the recognition model to obtain a dynamic terrain scoring result; inputting weather and vehicle working condition data into a recognition model to obtain a weight coefficient; according to the scheme, a vision and dynamics fusion method is used for identifying the terrain, and the defect that a single information source scheme is limited by weather or working conditions is overcome. Visual information and dynamic information are subjected to weighted fusion, weight coefficients are obtained through training according to factors such as weather and working conditions, the adaptive capacity is achieved, the difficulty of manually setting threshold values and weights is avoided, and the misrecognition proportion is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to a terrain recognition method, system, electronic device, and computer-readable medium. Background Technology

[0002] Off-road vehicles operate in complex environments, requiring mode switching to maintain performance in different scenarios. Manual operation is cumbersome and demands drivers' ability to accurately and promptly assess terrain, resulting in a poor driving experience. However, with the development of intelligent driving technology, technologies that automatically identify terrain using images, vehicle conditions, and map navigation are becoming increasingly practical. This technology can detect road surface types (such as ice, snow, mud, and gravel) or terrain features (such as slope and roughness) in real time, automatically adjusting parameters such as power distribution, suspension stiffness, and ESP intervention strategies to optimize handling and energy efficiency. Current implementation solutions can be categorized into three types: first, image sensor classification, using cameras, radar, or lidar to identify road textures or obstacles, which is susceptible to weather interference; second, dynamic data detection, using wheel speed sensors and inertial navigation to monitor wheel slippage or vehicle vibration frequency, indirectly inferring terrain (such as Mercedes-Benz's G-ROCS system), which suffers from response lag; and third, multi-source data fusion solutions, which combine data from multiple sources to improve accuracy, but this technology is complex and immature. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art, and proposes a terrain recognition method and system.

[0004] In a first aspect, embodiments of the present invention provide a terrain identification method, comprising:

[0005] Acquire visual data, dynamic data, and weather and vehicle operating condition data;

[0006] Visual data is input into the recognition model to obtain visual terrain scoring results.

[0007] Input the dynamic data into the recognition model to obtain the dynamic terrain score;

[0008] Input weather and vehicle operating condition data into the recognition model to obtain weight coefficients;

[0009] The visual and dynamic scoring results are fused based on weighting coefficients to output a fused terrain result.

[0010] In some embodiments, the dynamic data includes road surface slope, rolling resistance, wheel bounce energy, wheel speed noise energy, slip ratio, and vertical acceleration.

[0011] In some embodiments, the recognition model includes a graph classification model and a radial basis function neural network;

[0012] Visual data is input into the image classification model of the recognition model to obtain visual terrain scoring results;

[0013] The dynamic data is input into the radial basis neural network of the recognition model to obtain the dynamic terrain score.

[0014] In some embodiments, the recognition model includes a fully connected neural network;

[0015] Weather and vehicle operating condition data are input into a fully connected neural network to train and obtain the weight coefficients of the visual scoring results and the dynamic scoring results.

[0016] The weather and vehicle operating data include brightness, temperature and humidity, vehicle speed, engine speed, gear, braking force, steering angle, lateral acceleration and longitudinal acceleration.

[0017] In some embodiments, it also includes:

[0018] Obtain the initial image classification model, radial basis function neural network, and initial fully connected neural network;

[0019] The initial image classification model is pre-trained;

[0020] The initial radial basis neural network is pre-trained;

[0021] The pre-trained image classification model, radial basis function neural network, and initial fully connected neural network are trained together to obtain the recognition model.

[0022] In some embodiments, it also includes:

[0023] The projected terrain results are obtained through visual data, and the projected terrain results are spatiotemporally aligned with the fused terrain results.

[0024] Verify the previewed terrain results and the fused terrain results, and output the final terrain results.

[0025] In some embodiments, the verification of the pre-aimed terrain result and the fused terrain result, outputting the final terrain result, includes:

[0026] When the visual confidence level is lower than the set value, the fused terrain results are maintained for time t1.

[0027] When the visual confidence level is higher than the set value, the consistency between the previewed terrain result and the fused terrain result is judged. If the two are consistent, the previewed terrain result is used to maintain the result for time t2; if the two are inconsistent, the fused terrain result is used to maintain the result for time t1.

[0028] In a second aspect, embodiments of the present invention provide a terrain recognition system configured to implement any of the methods described above, including:

[0029] The data acquisition unit is used to acquire visual data, dynamic data, and weather and vehicle operating condition data;

[0030] The recognition unit obtains visual terrain scoring results based on visual data; obtains dynamic terrain scoring results based on dynamic data; obtains weighting coefficients based on weather and vehicle operating condition data; and fuses the visual scoring results and dynamic scoring results based on the weighting coefficients to output the fused terrain result.

[0031] Thirdly, the present invention also provides an electronic device, comprising:

[0032] One or more processors;

[0033] Memory, used to store one or more programs;

[0034] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods.

[0035] Fourthly, the present invention also provides a computer-readable medium on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps in any of the methods described.

[0036] The terrain recognition method provided by this invention includes acquiring visual data, dynamic data, and weather and vehicle operating condition data; inputting the visual data into a recognition model to obtain a visual terrain scoring result, and inputting the dynamic data into a recognition model to obtain a dynamic terrain scoring result; inputting the weather and vehicle operating condition data into the recognition model to obtain weight coefficients; and fusing the visual scoring result and the dynamic scoring result based on the weight coefficients to output a fused terrain result. This scheme uses a visual and dynamic fusion method to identify terrain, avoiding the shortcomings of single-information-source schemes that are limited by weather or operating conditions. Visual and dynamic information are fused using a weighted method, and the weight coefficients are trained based on factors such as weather and operating conditions, possessing adaptive capabilities, avoiding the difficulty of manually setting thresholds and weights, and reducing the misidentification rate. When training the network, the visual scoring and dynamic scoring modules are pre-trained separately first to obtain initial parameters before training the complete weight adaptive network, which speeds up the training speed and improves the accuracy of the final recognition result. A terrain buffer time is set to avoid the problem of result jumps, and the visually predicted terrain and the fused recognition result are verified, thereby speeding up the recognition result while ensuring accuracy. Attached Figure Description

[0037] Figure 1 This is a schematic diagram illustrating the steps of an embodiment of the terrain recognition method of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the steps of an embodiment of the present invention for spatiotemporal alignment of fused data;

[0039] Figure 3 This is a schematic diagram illustrating the steps of an embodiment of the weighted adaptive learning neural network structure of the present invention;

[0040] Figure 4 This is an overall architecture diagram of the terrain recognition method of the present invention;

[0041] Figure 5 This is a schematic diagram of the steps of the terrain pattern buffer judgment module of the present invention;

[0042] Figure 6 This is a schematic diagram of the structure of an embodiment of the terrain recognition system of the present invention;

[0043] Figure 7 This is a schematic diagram of the structure of an embodiment of the electronic device of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the technical solutions of the present invention, exemplary embodiments of the present invention are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0045] Where there is no conflict, the various embodiments of the present invention and the features thereof may be combined with each other.

[0046] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0047] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Terms such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0048] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having the meaning consistent with their meaning in the context of the relevant art and the invention, and will not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0049] In the technical solution of this invention, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information all comply with relevant laws and regulations and do not violate public order and good morals. The use of user data in this technical solution follows relevant national laws and regulations (e.g., the "Information Security Technology - Personal Information Security Specification"). For example: appropriate measures are taken for personal information access control; restrictions are imposed on the display of personal information; the purpose of using personal information does not exceed the scope of direct or reasonable association; and explicit identity targeting is eliminated when using personal information to avoid precisely locating a specific individual.

[0050] Among related technologies, pure image vision methods are easily affected by weather, pure dynamic methods are subject to lag and operating conditions, and fusion solutions are technically complex and costly.

[0051] To address at least one of the technical problems existing in the aforementioned related technologies, the present invention provides a terrain recognition method. Figure 1 The flowchart illustrates the steps of a terrain recognition method provided in an embodiment of the present invention.

[0052] like Figure 1 As shown, the terrain recognition method includes the following steps:

[0053] Step S10: Acquire visual data, dynamic data, and weather and vehicle operating condition data.

[0054] Specifically, visual data acquisition involves continuously capturing images of the area in front of and around the vehicle using onboard cameras (one or more, such as wide-angle, fisheye, or stereo cameras). The image data is transmitted via a vehicle network (such as Ethernet or a dedicated video cable) to a processing unit (such as an embedded GPU or a dedicated processor). During transmission, each frame is timestamped using hardware (e.g., using GPS time or a clock synchronized with the vehicle network).

[0055] Specifically, dynamics data acquisition involves acquiring sensor data related to vehicle dynamics via the vehicle's CAN bus. This includes: wheel speed sensors collecting the rotational speeds of the four wheels; an inertial measurement unit (IMU) collecting triaxial acceleration (including vertical acceleration) and triaxial angular velocity; suspension travel sensors collecting the suspension displacements of the four wheels; a steering angle sensor collecting the steering wheel angle; engine / motor torque data collecting drive torque information; and brake pressure data collecting brake system pressure. Similarly, each CAN signal carries a timestamp (either a timestamp from the CAN bus itself or a timestamp added upon reception).

[0056] Specifically, it involves acquiring weather and vehicle operating condition data.

[0057] For weather data, current environmental information is obtained through onboard weather sensors (such as light sensors, rain sensors, and temperature sensors). If there are no onboard sensors, real-time weather information (such as visibility and rainfall) can be obtained from meteorological services through vehicle networking functions (such as 4G / 5G).

[0058] For vehicle operating data, the following data is obtained from the CAN bus: vehicle speed, steering angle, braking status (brake pedal position or braking pressure), gear information (current gear of the vehicle, such as forward or reverse), and lighting status, including whether the headlights and fog lights are on (which can indirectly reflect weather conditions).

[0059] Visual, dynamic, and weather / vehicle condition data are collected in parallel, constituting all the data perceived by the system. All data sources must be synchronized via hardware and a global clock to be stamped with a unified and accurate timestamp. This is an absolute prerequisite for subsequent spatiotemporal alignment. After spatiotemporal alignment and feature extraction (dynamic index estimation, visual CNN feature extraction), the acquired raw data, along with weather / vehicle condition data, is input into a weighted adaptive learning neural network to ultimately complete terrain recognition.

[0060] Specifically, the sensor acquires images of the front and surrounding areas of the vehicle. The sensor achieves global clock synchronization through an agreed communication protocol and adds a hardware timestamp to each data packet. For data with different sampling frequencies, interpolation methods or interval sampling are used to achieve frequency synchronization. The sensor data is transformed from the vehicle coordinate system to the geodetic coordinate system so that the visual and dynamic data of the images are consistent in time and space.

[0061] It should be noted that this embodiment employs spatiotemporal alignment of dynamics-vision fusion data. The vehicle-mounted camera captures images of the vehicle's direction of travel and its surroundings, while the dynamic excitation comes from the tires, representing the vehicle's current position. There is a temporal and spatial misalignment between the two. Furthermore, the data frequencies collected by different sensors also differ. Spatiotemporal alignment is required before the fusion and discrimination algorithm to synchronize the frequencies of all data sources and reflect the terrain of the same area. This embodiment employs both hardware synchronization and software compensation synchronization. A pre-agreed communication protocol is used to achieve global clock synchronization. Additionally, hardware timestamps are added to each sensor to ensure data clock traceability. Data at different rates is aligned using interval sampling or interpolation methods, and spatially, the data from the vehicle coordinate system is transformed to the geodetic coordinate system.

[0062] It is understood that this embodiment uses spatiotemporal alignment of dynamic-visual fusion data, employing both hardware alignment and software compensation to ensure that the data sources for fusion discrimination represent the same region.

[0063] Taking the forward movement of a vehicle as an example, Figure 2 The road image captured by the sensor is in front of the vehicle, and its timestamp is... Assuming the vehicle acceleration remains unchanged over a short period, the timestamp of the dynamics data aligned with the visual data is: When the vehicle acceleration is large, the vehicle speed can be interpolated.

[0064] Based on the data, obtain dynamic indicators and output a dynamic feature vector consistent with the current terrain.

[0065] In this embodiment, in the step of obtaining dynamic indicators based on the data and outputting a dynamic feature vector consistent with the current terrain discrimination, the reference indicators for terrain discrimination include road surface slope, rolling resistance, wheel bounce energy, wheel speed noise energy, slip ratio, and vertical acceleration.

[0066] It is understandable that the main reference indicators for terrain identification include road surface slope, rolling resistance, wheel bounce energy, wheel speed noise energy, slip ratio, and vertical acceleration. Among these, only vertical acceleration can be obtained through sensor measurement, while the other parameters need to be estimated based on the corresponding models. Currently, there are mature algorithms available for all of these, so they will not be elaborated further.

[0067] Step S20: Input visual data into the recognition model to obtain visual terrain scoring results; input dynamic data into the recognition model to obtain dynamic terrain scoring results; input weather and vehicle operating condition data into the recognition model to obtain weight coefficients.

[0068] Furthermore, the recognition model includes a graphic classification model and a radial basis function neural network; visual data is input into the image classification model of the recognition model to obtain a visual terrain scoring result; dynamic data is input into the radial basis function neural network of the recognition model to obtain a dynamic terrain scoring result.

[0069] Furthermore, the aforementioned recognition model includes a fully connected neural network; weather and vehicle operating condition data are input into the fully connected neural network, and the weight coefficients of the visual scoring result and the dynamic scoring result are obtained through training; the weather and vehicle operating condition data include brightness, temperature and humidity, vehicle speed, RPM, gear, braking force, steering angle, lateral acceleration and longitudinal acceleration.

[0070] In this embodiment, it also includes:

[0071] The initial image classification model, radial basis function neural network, and initial fully connected neural network are obtained.

[0072] The initial image classification model is pre-trained.

[0073] The initial radial basis neural network is pre-trained.

[0074] The pre-trained image classification model, radial basis function neural network, and initial fully connected neural network are trained together to obtain the recognition model.

[0075] The process of inputting visual data into a recognition model to obtain visual terrain scoring results includes:

[0076] Specifically, visual data is processed by spatiotemporal alignment and preprocessing (e.g., distortion correction, color normalization) of single-frame images or short temporal images. Image pixel values ​​are normalized to the range [0,1] or [-1,1] and scaled to a fixed size required by the network (e.g., 224x224). A pre-trained convolutional neural network, such as ResNet, EfficientNet, or MobileNet, is used as the backbone network. This leverages general features learned on large datasets (e.g., ImageNet), accelerating convergence and improving performance.

[0077] Furthermore, convolutional neural networks include:

[0078] Input layer: Adapts to the image size and number of channels mentioned above.

[0079] Output layer: Replace the last fully connected layer of the CNN with a new layer whose number of neurons equals the number of terrains to be classified.

[0080] Activation function: The Softmax activation function is used in the output layer to normalize the output values ​​into a probability distribution. Each value represents the score (confidence) of the corresponding terrain, and the sum of all scores is 1.

[0081] The output of the convolutional neural network described above is a vector S_v=[s_v1,s_v2,...,s_vN] of length N (the number of terrain categories). Here, s_v1 represents the model's score that it considers the current image to be the first type of terrain (such as "asphalt road"), and its value is between [0,1].

[0082] The process involves inputting dynamic data into the recognition model to obtain dynamic terrain scoring results, including:

[0083] Specifically, the 6-dimensional dynamic feature vector [slope, rolling resistance, wheel hop energy, wheel speed noise energy, slip ratio, vertical acceleration] obtained after spatiotemporal alignment and index estimation is standardized to have a mean of 0 and a variance of 1.

[0084] In this embodiment, a standard RBF neural network is used, including:

[0085] Input layer: 6 neurons, corresponding to 6 dynamic features.

[0086] The hidden layer includes:

[0087] Number of neurons: M, usually M>6, the specific number is determined experimentally.

[0088] Kernel function: Each neuron uses a Gaussian radial basis function. The output of the i-th neuron is:

[0089]

[0090] in These are the center location and width parameters to be trained. Each neuron in the hidden layer is sensitive to a local region of the input space (controlled by the center point and width parameters) and is robust to input noise, making it suitable for terrain classification using range features of multiple dynamic metrics. The network has 6 input features, and the output size equals the number of terrain features to be classified.

[0091] The output layer consists of N neurons (the number of terrain categories). It is typically a linear layer, i.e., a weighted sum of the outputs from the hidden layers to the output layer. Finally, it is normalized to a probability distribution using the Softmax function.

[0092] The output of the RBF neural network described above is a vector of length N: S_k = [s_k1, s_k2, ..., s_kN]. Here, s_k1 represents the model's score indicating that the current dynamic feature belongs to type 1 terrain.

[0093] Specifically, weather and vehicle operating condition data are input into the recognition model to obtain weight coefficients, including:

[0094] In this embodiment, context feature vectors are obtained from sensors and the CAN bus. All continuous features are normalized, and discrete states are one-hot encoded.

[0095] Specifically, the context feature vector includes:

[0096] Weather-related factors: ambient light intensity, rain sensor signals, and image-based visibility estimates.

[0097] Vehicle operating condition related: vehicle speed, lateral acceleration, brake pedal opening, steering angular velocity, ABS / ESC activation status (0 / 1).

[0098] It is understandable that the above three steps can be executed in parallel in actual deployment to improve system efficiency. The parameters of all these modules are learned together through supervised training (pre-training + joint training) as described in the weight learning method. The outputs S_v, S_k, [w_v, w_k] of these three modules are fed into the weighted averaging stage to calculate the final fusion score: S_final = w_v * S_v + w_k * S_k. The terrain category with the highest score is the current fusion recognition result. In this way, the system not only integrates multimodal information but also dynamically adjusts the trust weights according to the real-time environment, thereby achieving robust and adaptive off-road terrain recognition.

[0099] Step S30: Based on the weighting coefficients, fuse the visual scoring results and the dynamic scoring results to output the fused terrain results.

[0100] Specifically, by combining the three modules mentioned above, a weighted average is calculated and normalized to the range of [0,1] to obtain the final terrain score. The neural network in this scheme is trained under supervision, and the labels are converted into One-Hot codes, i.e., the corresponding terrain is assigned a 1 and other terrains are assigned a 0, and the cross-entropy loss function is used.

[0101] The specific training steps are as follows:

[0102] Pre-training: The visual scoring module and the dynamics scoring module are trained separately first. When training the visual scoring module, visual data is used as input, and the corresponding labels are used as targets; the neural network is trained under supervised supervision. When training the dynamics scoring module, dynamic data is used as input, and the same labels are used as targets; the neural network is trained under supervised supervision.

[0103] Overall training: The three modules are trained jointly, with a focus on training the parameters of the weight adaptive module.

[0104] It's understandable that during overall training, the input to the weight adaptation module is weather and vehicle condition data, which are not directly related to terrain categories but rather to the reliability of each module. Therefore, through joint training, the network learns to adjust the weights based on this contextual information, resulting in a more accurate fused result.

[0105] Furthermore, during training, we use gradient descent algorithms (such as Adam) to minimize the cross-entropy loss.

[0106] Since the fusion score S_fusion is a function of the weights output by the weight adaptation module, and the weights are in turn a function of the weather and vehicle operating condition data, the parameters of the weight adaptation module can be updated by backpropagating the error using the chain rule.

[0107] Please see Figure 4 Pure vision methods can predict the terrain in the direction of vehicle travel. Therefore, under appropriate conditions, pure vision classification results can be used to speed up terrain identification. The main part uses a weighted adaptive method to fuse visual and dynamic information, taking the item with the highest score as the fused terrain recognition result. On the other hand, the pure vision scheme obtains the pre-targeted terrain result, caches it, and then... Figure 2 The alignment calculation method shown extracts the result at the same location as the fused terrain representation; the two results are input into the terrain pattern buffer judgment module to obtain the final identified terrain.

[0108] Furthermore, it also includes: obtaining the pre-aimed terrain result through visual data, and aligning the pre-aimed terrain result with the fused terrain result in time and space; verifying the pre-aimed terrain result and the fused terrain result, and outputting the final terrain result.

[0109] It should be noted that: the pre-aimed terrain obtained purely from vision (the sensor collects images of the front and surrounding areas of the vehicle) is cached, the sensor achieves global clock synchronization through an agreed communication protocol, and adds a hardware timestamp to each data packet. For data with different sampling frequencies, interpolation methods or interval sampling are used to achieve frequency synchronization. The sensor data is transformed from the vehicle coordinate system to the geodetic coordinate system so that the visual and dynamic data of the image are consistent in time and space.

[0110] The verification of the pre-aimed terrain results and the fused terrain results outputs the final terrain results, including:

[0111] When the visual confidence level is lower than the set value, the fused terrain results are maintained for time t1.

[0112] When the visual confidence level is higher than the set value, the consistency between the previewed terrain result and the fused terrain result is judged. If the two are consistent, the previewed terrain result is used to maintain the result for time t2; if the two are inconsistent, the fused terrain result is used to maintain the result for time t1.

[0113] Specifically, please refer to Figure 5 The step of inputting the target terrain and the fused terrain into the terrain pattern buffer judgment module to obtain the final identified terrain includes:

[0114] Input the target terrain and the merged terrain;

[0115] Determine if the visual confidence level is high; if yes, proceed to the next step; if no, determine if the fused terrain has been maintained for time t1; if yes, switch the terrain; if no, do not switch the terrain.

[0116] Determine whether the target terrain and the merged terrain are consistent; if yes, proceed to the next step; if no, determine whether the merged terrain has been maintained for time t1; if yes, switch terrain; if no, do not switch terrain.

[0117] Determine whether the consistent state has been maintained for time t2; if so, proceed to the next step; determine whether the merged terrain has been maintained for time t1; if so, switch the terrain; if not, do not switch the terrain.

[0118] Switch terrain.

[0119] It is understood that this embodiment uses a terrain pattern buffer judgment module to set two levels of buffer time to avoid changes in terrain recognition results. At the same time, it verifies the results of the pre-aimed terrain and the fused terrain, ensuring recognition accuracy while speeding up the recognition process.

[0120] It should be noted that the terrain recognition program requires a certain amount of time to calculate, and a buffer time is needed to ensure the stability of the recognition result. This embodiment sets two levels of buffer time based on information verification capability, where t1 > t2. When the visual confidence is not high enough (e.g., below 80%), the fused terrain recognition result is detected, maintained for time t1 (e.g., 3s), and then the final result is switched. When the visual confidence is high, the pure visual discrimination result is verified with the fused recognition result. If they match, the buffer time is shortened to t2 (e.g., 0.5s), thereby accelerating the recognition speed through multi-source data fusion.

[0121] The terrain recognition method provided by this invention includes acquiring visual data, dynamic data, and weather and vehicle operating condition data; inputting the visual data into a recognition model to obtain a visual terrain scoring result, and inputting the dynamic data into a recognition model to obtain a dynamic terrain scoring result; inputting the weather and vehicle operating condition data into the recognition model to obtain weight coefficients; and fusing the visual scoring result and the dynamic scoring result based on the weight coefficients to output a fused terrain result. This scheme uses a visual and dynamic fusion method to identify terrain, avoiding the shortcomings of single-information-source schemes that are limited by weather or operating conditions. Visual and dynamic information are fused using a weighted method, and the weight coefficients are trained based on factors such as weather and operating conditions, possessing adaptive capabilities, avoiding the difficulty of manually setting thresholds and weights, and reducing the misidentification rate. When training the network, the visual scoring and dynamic scoring modules are pre-trained separately first to obtain initial parameters before training the complete weight adaptive network, which speeds up the training speed and improves the accuracy of the final recognition result. A terrain buffer time is set to avoid the problem of result jumps, and the visually predicted terrain and the fused recognition result are verified, thereby speeding up the recognition result while ensuring accuracy.

[0122] Please see Figure 6 The present invention also provides a terrain recognition system. It is applied to the terrain recognition method provided in the above embodiments and specifically includes: a data acquisition unit and a recognition unit.

[0123] The data acquisition unit is used to acquire visual data, dynamic data, and weather and vehicle operating condition data.

[0124] Specifically, visual data acquisition involves continuously capturing images of the area in front of and around the vehicle using onboard cameras (one or more, such as wide-angle, fisheye, or stereo cameras). The image data is transmitted via a vehicle network (such as Ethernet or a dedicated video cable) to a processing unit (such as an embedded GPU or a dedicated processor). During transmission, each frame is timestamped using hardware (e.g., using GPS time or a clock synchronized with the vehicle network).

[0125] Specifically, dynamics data acquisition involves acquiring sensor data related to vehicle dynamics via the vehicle's CAN bus. This includes: wheel speed sensors collecting the rotational speeds of the four wheels; an inertial measurement unit (IMU) collecting triaxial acceleration (including vertical acceleration) and triaxial angular velocity; suspension travel sensors collecting the suspension displacements of the four wheels; a steering angle sensor collecting the steering wheel angle; engine / motor torque data collecting drive torque information; and brake pressure data collecting brake system pressure. Similarly, each CAN signal carries a timestamp (either a timestamp from the CAN bus itself or a timestamp added upon reception).

[0126] Specifically, it involves acquiring weather and vehicle operating condition data.

[0127] For weather data, current environmental information is obtained through onboard weather sensors (such as light sensors, rain sensors, and temperature sensors). If there are no onboard sensors, real-time weather information (such as visibility and rainfall) can be obtained from meteorological services through vehicle networking functions (such as 4G / 5G).

[0128] For vehicle operating data, the following data is obtained from the CAN bus: vehicle speed, steering angle, braking status (brake pedal position or braking pressure), gear information (current gear of the vehicle, such as forward or reverse), and lighting status, including whether the headlights and fog lights are on (which can indirectly reflect weather conditions).

[0129] Visual, dynamic, and weather data are collected in parallel, constituting all the data perceived by the system. All data sources must be synchronized via hardware and a global clock to ensure a unified and accurate timestamp. This is an absolute prerequisite for subsequent spatiotemporal alignment. After spatiotemporal alignment and feature extraction (dynamic index estimation and visual CNN feature extraction), the acquired raw data, along with weather and weather data, are input into a weighted adaptive learning neural network to ultimately complete terrain recognition.

[0130] Specifically, the sensor acquires images of the front and surrounding areas of the vehicle. The sensor achieves global clock synchronization through an agreed communication protocol and adds a hardware timestamp to each data packet. For data with different sampling frequencies, interpolation methods or interval sampling are used to achieve frequency synchronization. The sensor data is transformed from the vehicle coordinate system to the geodetic coordinate system so that the visual and dynamic data of the images are consistent in time and space.

[0131] It should be noted that this embodiment employs spatiotemporal alignment of dynamics-vision fusion data. The vehicle-mounted camera captures images of the vehicle's direction of travel and its surroundings, while the dynamic excitation comes from the tires, representing the vehicle's current position. There is a temporal and spatial misalignment between the two. Furthermore, the data frequencies collected by different sensors also differ. Spatiotemporal alignment is required before the fusion and discrimination algorithm to synchronize the frequencies of all data sources and reflect the terrain of the same area. This embodiment employs both hardware synchronization and software compensation synchronization. A pre-agreed communication protocol is used to achieve global clock synchronization. Additionally, hardware timestamps are added to each sensor to ensure data clock traceability. Data at different rates is aligned using interval sampling or interpolation methods, and spatially, the data from the vehicle coordinate system is transformed to the geodetic coordinate system.

[0132] It is understood that this embodiment uses spatiotemporal alignment of dynamic-visual fusion data, employing both hardware alignment and software compensation to ensure that the data sources for fusion discrimination represent the same region.

[0133] The recognition unit obtains visual terrain scoring results based on visual data; obtains dynamic terrain scoring results based on dynamic data; obtains weighting coefficients based on weather and vehicle operating condition data; and fuses the visual scoring results and dynamic scoring results based on the weighting coefficients to output the fused terrain result.

[0134] Furthermore, the recognition model includes a graphic classification model and a radial basis function neural network; visual data is input into the image classification model of the recognition model to obtain a visual terrain scoring result; dynamic data is input into the radial basis function neural network of the recognition model to obtain a dynamic terrain scoring result.

[0135] Furthermore, the aforementioned recognition model includes a fully connected neural network; weather and vehicle operating condition data are input into the fully connected neural network, and the weight coefficients of the visual scoring result and the dynamic scoring result are obtained through training; the weather and vehicle operating condition data include brightness, temperature and humidity, vehicle speed, RPM, gear, braking force, steering angle, lateral acceleration and longitudinal acceleration.

[0136] In this embodiment, it also includes:

[0137] The initial image classification model, radial basis function neural network, and initial fully connected neural network are obtained.

[0138] The initial image classification model is pre-trained.

[0139] The initial radial basis neural network is pre-trained.

[0140] The pre-trained image classification model, radial basis function neural network, and initial fully connected neural network are trained together to obtain the recognition model.

[0141] The process of inputting visual data into a recognition model to obtain visual terrain scoring results includes:

[0142] Specifically, visual data is processed by spatiotemporal alignment and preprocessing (e.g., distortion correction, color normalization) of single-frame images or short temporal images. Image pixel values ​​are normalized to the range [0,1] or [-1,1] and scaled to a fixed size required by the network (e.g., 224x224). A pre-trained convolutional neural network, such as ResNet, EfficientNet, or MobileNet, is used as the backbone network. This leverages general features learned on large datasets (e.g., ImageNet), accelerating convergence and improving performance.

[0143] Furthermore, convolutional neural networks include:

[0144] Input layer: Adapts to the image size and number of channels mentioned above.

[0145] Output layer: Replace the last fully connected layer of the CNN with a new layer whose number of neurons equals the number of terrains to be classified.

[0146] Activation function: The Softmax activation function is used in the output layer to normalize the output values ​​into a probability distribution. Each value represents the score (confidence) of the corresponding terrain, and the sum of all scores is 1.

[0147] The output of the convolutional neural network described above is a vector S_v=[s_v1,s_v2,...,s_vN] of length N (the number of terrain categories). Here, s_v1 represents the model's score that it considers the current image to be the first type of terrain (such as "asphalt road"), and its value is between [0,1].

[0148] The process involves inputting dynamic data into the recognition model to obtain dynamic terrain scoring results, including:

[0149] Specifically, the 6-dimensional dynamic feature vector [slope, rolling resistance, wheel hop energy, wheel speed noise energy, slip ratio, vertical acceleration] obtained after spatiotemporal alignment and index estimation is standardized to have a mean of 0 and a variance of 1.

[0150] In this embodiment, a standard RBF neural network is used, including:

[0151] Input layer: 6 neurons, corresponding to 6 dynamic features.

[0152] The hidden layer includes:

[0153] Number of neurons: M, usually M>6, the specific number is determined experimentally.

[0154] Kernel function: Each neuron uses a Gaussian radial basis function. The output of the i-th neuron is:

[0155]

[0156] in These are the center location and width parameters to be trained. Each neuron in the hidden layer is sensitive to a local region of the input space (controlled by the center point and width parameters) and is robust to input noise, making it suitable for terrain classification using range features of multiple dynamic metrics. The network has 6 input features, and the output size equals the number of terrain features to be classified.

[0157] The output layer consists of N neurons (the number of terrain categories). It is typically a linear layer, i.e., a weighted sum of the outputs from the hidden layers to the output layer. Finally, it is normalized to a probability distribution using the Softmax function.

[0158] The output of the RBF neural network described above is a vector of length N: S_k = [s_k1, s_k2, ..., s_kN]. Here, s_k1 represents the model's score indicating that the current dynamic feature belongs to type 1 terrain.

[0159] Specifically, weather and vehicle operating condition data are input into the recognition model to obtain weight coefficients, including:

[0160] In this embodiment, context feature vectors are obtained from sensors and the CAN bus. All continuous features are normalized, and discrete states are one-hot encoded.

[0161] Specifically, the context feature vector includes:

[0162] Weather-related factors: ambient light intensity, rain sensor signals, and image-based visibility estimates.

[0163] Vehicle operating condition related: vehicle speed, lateral acceleration, brake pedal opening, steering angular velocity, ABS / ESC activation status (0 / 1).

[0164] It is understandable that the above three steps can be executed in parallel in actual deployment to improve system efficiency. The parameters of all these modules are learned together through supervised training (pre-training + joint training) as described in the weight learning method. The outputs S_v, S_k, [w_v, w_k] of these three modules are fed into the weighted averaging stage to calculate the final fusion score: S_final = w_v * S_v + w_k * S_k. The terrain category with the highest score is the current fusion recognition result. In this way, the system not only integrates multimodal information but also dynamically adjusts the trust weights according to the real-time environment, thereby achieving robust and adaptive off-road terrain recognition.

[0165] Step S30: Based on the weighting coefficients, fuse the visual scoring results and the dynamic scoring results to output the fused terrain results.

[0166] Specifically, by combining the three modules mentioned above, a weighted average is calculated and normalized to the range of [0,1] to obtain the final terrain score. The neural network in this scheme is trained under supervision, and the labels are converted into One-Hot codes, i.e., the corresponding terrain is assigned a 1 and other terrains are assigned a 0, and the cross-entropy loss function is used.

[0167] The specific training steps are as follows:

[0168] Pre-training: The visual scoring module and the dynamics scoring module are trained separately first. When training the visual scoring module, visual data is used as input, and the corresponding labels are used as targets; the neural network is trained under supervised supervision. When training the dynamics scoring module, dynamic data is used as input, and the same labels are used as targets; the neural network is trained under supervised supervision.

[0169] Overall training: The three modules are trained together, with a focus on training the parameters of the weight adaptive module.

[0170] It's understandable that during overall training, the input to the weight adaptation module is weather and vehicle condition data, which are not directly related to terrain categories but rather to the reliability of each module. Therefore, through joint training, the network learns to adjust the weights based on this contextual information, resulting in a more accurate fused result.

[0171] Furthermore, during training, we use gradient descent algorithms (such as Adam) to minimize the cross-entropy loss.

[0172] Since the fusion score S_fusion is a function of the weights output by the weight adaptation module, and the weights are in turn a function of the weather and vehicle operating condition data, the parameters of the weight adaptation module can be updated by backpropagating the error using the chain rule.

[0173] Furthermore, it also includes: obtaining the pre-aimed terrain result through visual data, and aligning the pre-aimed terrain result with the fused terrain result in time and space; verifying the pre-aimed terrain result and the fused terrain result, and outputting the final terrain result.

[0174] It should be noted that: the pre-aimed terrain obtained purely from vision (the sensor collects images of the front and surrounding areas of the vehicle) is cached, the sensor achieves global clock synchronization through an agreed communication protocol, and adds a hardware timestamp to each data packet. For data with different sampling frequencies, interpolation methods or interval sampling are used to achieve frequency synchronization. The sensor data is transformed from the vehicle coordinate system to the geodetic coordinate system so that the visual and dynamic data of the image are consistent in time and space.

[0175] The verification of the pre-aimed terrain results and the fused terrain results outputs the final terrain results, including:

[0176] When the visual confidence level is lower than the set value, the fused terrain results are maintained for time t1.

[0177] When the visual confidence level is higher than the set value, the consistency between the previewed terrain result and the fused terrain result is judged. If the two are consistent, the previewed terrain result is used to maintain the result for time t2; if the two are inconsistent, the fused terrain result is used to maintain the result for time t1.

[0178] Specifically, please refer to Figure 5 The step of inputting the target terrain and the fused terrain into the terrain pattern buffer judgment module to obtain the final identified terrain includes:

[0179] Input the target terrain and the merged terrain;

[0180] Determine if the visual confidence level is high; if yes, proceed to the next step; if no, determine if the fused terrain has been maintained for time t1; if yes, switch the terrain; if no, do not switch the terrain.

[0181] Determine whether the target terrain and the merged terrain are consistent; if yes, proceed to the next step; if no, determine whether the merged terrain has been maintained for time t1; if yes, switch terrain; if no, do not switch terrain.

[0182] Determine whether the consistent state has been maintained for time t2; if so, proceed to the next step; determine whether the merged terrain has been maintained for time t1; if so, switch the terrain; if not, do not switch the terrain.

[0183] Switch terrain.

[0184] It is understood that this embodiment uses a terrain pattern buffer judgment module to set two levels of buffer time to avoid changes in terrain recognition results. At the same time, it verifies the results of the pre-aimed terrain and the fused terrain, ensuring recognition accuracy while speeding up the recognition process.

[0185] It should be noted that the terrain recognition program requires a certain amount of time to calculate, and a buffer time is needed to ensure the stability of the recognition result. This embodiment sets two levels of buffer time based on information verification capability, where t1 > t2. When the visual confidence is not high enough (e.g., below 80%), the fused terrain recognition result is detected, maintained for time t1 (e.g., 3s), and then the final result is switched. When the visual confidence is high, the pure visual discrimination result is verified with the fused recognition result. If they match, the buffer time is shortened to t2 (e.g., 0.5s), thereby accelerating the recognition speed through multi-source data fusion.

[0186] The terrain recognition system provided by this invention includes acquiring visual data, dynamic data, and weather and vehicle operating condition data; inputting visual data into a recognition model to obtain a visual terrain scoring result, and inputting dynamic data into a recognition model to obtain a dynamic terrain scoring result; inputting weather and vehicle operating condition data into the recognition model to obtain weight coefficients; and fusing the visual scoring result and the dynamic scoring result based on the weight coefficients to output a fused terrain result. This solution uses a visual and dynamic fusion method to identify terrain, avoiding the shortcomings of single-information-source solutions that are limited by weather or operating conditions. Visual and dynamic information are fused using a weighted method, and the weight coefficients are trained based on factors such as weather and operating conditions, possessing adaptive capabilities and avoiding the difficulty of manually setting thresholds and weights, thus reducing the misidentification rate. When training the network, the visual scoring and dynamic scoring modules are pre-trained separately to obtain initial parameters before training the complete weighted adaptive network, accelerating the training speed and improving the accuracy of the final recognition result. A terrain buffer time is set to avoid the problem of result jumps, and the visually predicted terrain and the fused recognition result are verified, thereby accelerating the recognition result while ensuring accuracy.

[0187] Based on the same inventive concept, embodiments of the present invention also provide an electronic device. Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Figure 7 As shown, an embodiment of the present invention provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement any of the terrain recognition methods described in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, configured to enable information interaction between the processor and the memory.

[0188] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus).

[0189] In some embodiments, the processor 101, memory 102, and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0190] In some embodiments, the one or more processors 101 include a field-programmable gate array.

[0191] This invention also provides a computer-readable medium. The computer-readable medium stores a computer program, which, when executed by a processor, implements the steps of any of the terrain recognition methods described in the above embodiments. The computer-readable storage medium may be volatile or non-volatile.

[0192] This invention also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described terrain recognition method.

[0193] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0194] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0195] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0196] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0197] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0198] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0199] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0200] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0201] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0202] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A terrain recognition method, characterized in that, It includes: Acquire visual data, dynamic data, and weather and vehicle operating condition data; Visual data is input into the recognition model to obtain visual terrain scoring results. Input the dynamic data into the recognition model to obtain the dynamic terrain score; Input weather and vehicle operating condition data into the recognition model to obtain weight coefficients; The visual and dynamic scoring results are fused based on weighting coefficients to output a fused terrain result.

2. The terrain recognition method according to claim 1, characterized in that, The dynamic data includes road surface slope, rolling resistance, wheel bounce energy, wheel speed noise energy, slip ratio, and vertical acceleration.

3. The terrain recognition method according to claim 1, characterized in that, The recognition model includes a graph classification model and a radial basis function neural network; Visual data is input into the image classification model of the recognition model to obtain visual terrain scoring results; The dynamic data is input into the radial basis neural network of the recognition model to obtain the dynamic terrain score.

4. The terrain identification method according to claim 1, characterized in that, The recognition model includes a fully connected neural network; Weather and vehicle operating condition data are input into a fully connected neural network to train and obtain the weight coefficients of the visual scoring results and the dynamic scoring results. The weather and vehicle operating data include brightness, temperature and humidity, vehicle speed, engine speed, gear, braking force, steering angle, lateral acceleration and longitudinal acceleration.

5. The terrain identification method according to claim 1, characterized in that, Also includes: Obtain the initial image classification model, radial basis function neural network, and initial fully connected neural network; The initial image classification model is pre-trained; The initial radial basis neural network is pre-trained; The pre-trained image classification model, radial basis function neural network, and initial fully connected neural network are trained together to obtain the recognition model.

6. The terrain identification method according to claim 1, characterized in that, Also includes: The projected terrain results are obtained through visual data, and the projected terrain results are spatiotemporally aligned with the fused terrain results. Verify the previewed terrain results and the fused terrain results, and output the final terrain results.

7. The terrain identification method according to claim 6, characterized in that, The verification of the pre-aimed terrain results and the fused terrain results outputs the final terrain results, including: When the visual confidence level is lower than the set value, the fused terrain results are maintained for time t1. When the visual confidence level is higher than the set value, the consistency between the previewed terrain result and the fused terrain result is judged. If the two are consistent, the previewed terrain result is used to maintain the result for time t2; if the two are inconsistent, the fused terrain result is used to maintain the result for time t1.

8. A terrain recognition system, characterized in that, Configured to implement the method of any one of claims 1-7, comprising: The data acquisition unit is used to acquire visual data, dynamic data, and weather and vehicle operating condition data; The recognition unit obtains visual terrain scoring results based on visual data; obtains dynamic terrain scoring results based on dynamic data; obtains weighting coefficients based on weather and vehicle operating condition data; and fuses the visual scoring results and dynamic scoring results based on the weighting coefficients to output the fused terrain result.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images