346 results about "Street scene" patented technology

The invention relates to the technical field of landscape evaluation, and discloses an urban green land landscape evaluation method based on a large model, and the method comprises the steps: carrying out the real-time analysis and standardization processing of a streetscape image, IoT environment data, satellite vegetation coverage data and RTK high-precision positioning information of a target city district, extracting core landscape elements, building cross-scene semantic mapping, and carrying out the calculation of the cross-scene semantic mapping. Transmitting the standardized data to a central database and deploying edge nodes; extracting a multi-dimensional landscape index by adopting an improved semantic segmentation model, dynamically adjusting an index weight in combination with regional features and seasonal changes, and classifying and correcting deviation data by edge nodes; a visual evaluation result is generated based on a cooperative computing architecture and a digital twinborn model, a landscape space to be promoted is identified, and optimization suggestions are generated; and obtaining planner feedback information, updating the model, the weight rule and the suggestion generation strategy, and forming a closed loop iteration mechanism. According to the method, the dynamic and practical evaluation requirements of the current urban green land landscape can be met.

The invention discloses a mobile terminal streetscape image real-time segmentation method based on a lightweight neural network, and relates to the technical field of image segmentation. The method comprises the following steps: firstly, carrying out 320 * 320 adjustment, Z-score standardization, adaptive histogram equalization and 3 * 3 Gaussian filtering preprocessing on an input streetscape image; then, an improved MobileNetV3 backbone network is used, and a five-scale feature map is output in combination with DropBlock regularization through eight feature extraction stages including depth separable convolution and an SE attention module; multi-scale features are fused through a U-shaped structure, and a fusion feature map is generated through up-sampling, element-by-element addition of dimension reduction low-layer features and an attention gating module; and during reasoning, outputting a segmentation mask by using a convolutional layer, Softmax and a conditional random field, and finally performing knowledge distillation, weight pruning, 8-bit quantization and TensorRT optimization. According to the invention, high-precision real-time street view segmentation is realized, the robustness is high, and the method is suitable for different devices and scenes.

The invention relates to the field of real-time visualization of road conditions, in particular to a data simulation visualization rendering method and system for urban traffic road conditions. The method comprises the following steps: extracting a real-time satellite streetscape image based on urban satellite remote sensing scanning, and performing scene pixel-level segmentation to obtain scene texture rendering parameters; scene illumination visual identification is carried out according to the real-time satellite streetscape image, traffic scene background modeling is carried out based on scene texture rendering parameters, and a real-time scene background model is constructed; the method comprises the following steps: acquiring urban-level multi-source traffic monitoring data flow, performing vehicle state sensing, and constructing a multi-dimensional particle feature matrix; road network topological correlation analysis and global traffic network state perception are carried out according to the real-time satellite streetscape images, and a road network state perception model is constructed. According to the invention, a real real-time traffic environment is visualized, scene effects in different traffic states are presented, the current road condition can be rapidly evaluated, and the traffic control decision efficiency is improved.

The invention discloses a time sequence streetscape image-based building update type identification method and device, and relates to the technical field of civil engineering and computer vision. The method comprises the following steps: acquiring a plurality of streetscape images; outputting a target detection result of the building through a YOLO target detection model; according to the metadata information of the streetscape image and the target detection result, determining geographic coordinates of the building; constructing an initial building group geographic information database by combining a target re-recognition algorithm according to the geographic coordinates; constructing a building update type identification data set according to the initial building group geographic information database; building a building update type identification model; using the building update type identification data set to train a building update type identification model; and outputting an update type identification result of the building through the trained building update type identification model. And storing the update type identification result to the initial building group geographic information database to obtain a target building group geographic information database.

The invention provides an element feature measurement method and device for block style and appearance shaping, which can be applied to the field of image processing and the field of urban and rural planning and design. The method comprises the steps that a trained semantic segmentation model is used for processing a streetscape image, an element type proportion and a segmented image are obtained, and the element type proportion represents the area proportion of elements with element types in the streetscape image; processing the streetscape image by using a large language model to obtain a streetscape description text; performing feature extraction on the streetscape description text to obtain streetscape semantic features; streetscape visual features and streetscape semantic features obtained by performing feature extraction on the element type proportions and the segmented images are fused, and multi-modal fusion features are obtained; and inputting the multi-modal fusion features into a trained evaluation model, and outputting an evaluation result, the evaluation result comprising a plurality of evaluation values used for measuring features of each element in the streetscape image for different evaluation dimensions, and the evaluation result being used for shaping block styles and features with endemic features.

The invention belongs to the technical field of city planning, and discloses a dynamic adjustment method for a city update planning scheme and related equipment, and the method comprises the steps: obtaining city street view video data through the normalized cruise shooting of an unmanned plane; preprocessing the urban street scene video data, identifying building facade damage information and traffic information from the preprocessed urban street scene video data, and evaluating a building facade damage level and a traffic influence level; based on a preset building facade damage grade threshold value, a traffic influence grade threshold value, a building facade damage risk weight and a traffic influence risk weight, constructing a weighted risk assessment model, and according to the building facade damage grade and the traffic influence grade, calculating a comprehensive risk score; determining an adjustment strategy according to the comprehensive risk score, so as to dynamically adjust the urban update planning scheme; therefore, the adjustment efficiency of the urban update planning scheme can be improved.

The embodiment of the invention discloses an address search method and device based on deep learning, a storage medium and a server, and relates to the field of logistics information processing. The method comprises the following steps: firstly, acquiring data such as text addresses, space coordinates, peripheral POIs (Point of Interest) and streetscape pictures from a multi-modal address database; text addresses are decomposed into different levels and converted into corresponding vectors, and then the vectors are spliced into text address vectors. The space coordinates, the POI and the streetscape pictures are also converted into vectors respectively, and the vectors are subjected to weighted summation to generate multi-modal vectors; meanwhile, the text address entity inquires the knowledge base to obtain semantic enhancement information, encodes the semantic enhancement information into a vector, splices the vector with the multi-modal vector to form a final address vector, and stores the final address vector into an address vector library. After the user inputs the query text address, the query vector is generated, the candidate address vector with high similarity is retrieved in the library, and detailed information of the candidate address is displayed on the interactive interface, so that the accuracy and richness of address description are improved, and an efficient address query function is realized.

A method and an apparatus for predicting a road traffic carbon emission based on panoramic images, a device and a medium are provided, relating to the technical field of data prediction. In the method, a historical street view image of an observation area is acquired from the Internet, feature analysis is performed on the historical street view image to obtain a historical feature vector, and a road traffic carbon emission predicted value is obtained based on the historical feature vector and a carbon emission prediction model. With the method or the apparatus, auxiliary explanations for carbon emission sources in cities can be provided based on features of street views.

The invention discloses a shielding complementing method based on a continuous streetscape panoramic image, and aims to solve the problem of information loss caused by shielding of dynamic objects (such as vehicles and pedestrians) in the existing streetscape image, and the method comprises the following steps: S1, recognizing and positioning a dynamic shielding object area in a continuous streetscape panoramic image sequence; s2, providing a continuous panoramic image-oriented feature extraction and matching algorithm (FDMPano), realizing robust feature retrieval and matching in a cross-view angle, and accurately positioning a reference image region which can be complemented; and S3, carrying out feature alignment and content mapping on the sheltered area based on a matching result, and generating an unsheltered panoramic image with consistent vision. According to the method, the internal relevance of continuous streetscape data is fully utilized, the occlusion area can be automatically completed with high quality, the integrity and availability of streetscape images are remarkably improved, and the method has wide application value in the fields of automatic driving, digital cities, virtual reality and the like.

According to the cold region city emotion perception evaluation and measurement method, a cold region city is selected as a research object, and physiological signals and subjective scores are comprehensively considered; street view images are scored by training the street view perception model, perception of the street view images is realized, physiological signals and subjective scores are combined to realize comprehensive evaluation of corresponding areas of the cold region city, and a comprehensive evaluation result is less subject to subjective influence and is stable; the method can analyze the influence of the winter street space of the cold region city on the residential emotion perception, accurately measure the quality of the cold region city street space influencing the residential emotion, has the advantages of high efficiency and easy popularization and application, can be widely applied to the field of urban planning living street space quality investigation and measurement, and has a wide application prospect. And a method and a basis are provided for cold region city planning street space optimization practice.

The invention discloses an environmental health risk assessment system, electronic equipment and a nonvolatile storage medium. The system comprises an environment image acquisition module which is used for acquiring an environment image of a target area; the image feature extraction module is used for performing feature extraction on the environment image to obtain a first feature corresponding to the streetscape image and a second feature corresponding to the remote sensing image; the environment variable prediction module is used for fusing the first feature and the second feature, and determining a parameter value of an environment variable corresponding to the target area according to a fused feature obtained after fusion; and the comprehensive index determination module is used for determining a comprehensive environment index corresponding to the target crowd object in the environment of the target area according to the parameter value of the environment variable. According to the method and the device, the technical problem that the environmental health risk cannot be effectively assessed due to the fact that a method for measuring and analyzing the environmental variables mainly depends on field monitoring, the efficiency is low and the data coverage is narrow in the prior art is solved.

The invention discloses a greenbelt cooling effect evaluation method and system based on a remote sensing image and a streetscape image, and the method comprises the steps: obtaining the data of the streetscape image, carrying out the denoising and cutting, recognizing, extracting and calculating the pixel coverage rate of arbors, grass trees and shrubs, and constructing the greening feature indexes of the streetscape image; acquiring an urban satellite remote sensing image, performing radiometric calibration, correction and identification, and constructing a greening characteristic index of the remote sensing image; acquiring a multispectral remote sensing image of the city, performing cloud detection, and obtaining the surface temperature of the city through surface temperature inversion; and constructing a random forest model and performing iterative training based on greening characteristic indexes of the surface temperature, the streetscape image and the remote sensing image of the city, and evaluating the cooling effect by using the fitted random forest model. The method gives full play to the advantages of wide coverage range of the remote sensing image and rich details of the street scene image, makes up for the limitation of a single data source, and achieves the comprehensive and accurate evaluation of the cooling effect of the green land.

The invention discloses an urban traffic three-dimensional model vegetation element automatic generation method and system based on streetscape recognition, a terminal and a storage medium. The method comprises the steps of constructing a vegetation element knowledge graph, defining a vegetation type and typical attributes, directionally collecting a three-dimensional model and a multi-view image, unifying metadata and storing the metadata. Pixel-level semantic segmentation is carried out on the streetscape image, vegetation categories are identified based on transfer learning and an attention mechanism, and ROI and feature vectors are output; screening the candidate set based on ontology mapping, selecting a best matched vegetation model for the local feature vector of the recognized ROI and a pre-stored feature vector in a material library, and recording mapping; based on reference object correction, knowledge graph fusion and automatic discrimination or optimization strategies, reliable scale correction, number or position deduction and batch three-dimensional model import placement meeting ecological constraints are carried out on the vegetation ROI. According to the invention, efficient, intelligent and standardized technical support is provided for traffic simulation and urban visualization.

The invention provides a signal system and method for urban street view visual emotion prediction and application. The signal system comprises a monitoring module, a processing module, a first construction module, a second construction module and an output module. The monitoring module is used for collecting electroencephalogram signals and emotion evaluation reaction indexes when the testee watches the streetscape image; the processing module obtains relative power of electroencephalogram signals of a single-image presentation section and a multi-image sequence section; the first construction module is used for constructing an electroencephalogram signal emotion prediction model by using multi-image sequence segment electroencephalogram signal relative power and emotion evaluation reaction indexes, and is applied to single-image presentation segment electroencephalogram signals to obtain an emotion reaction of any streetscape image; the second construction module is used for constructing a machine learning model, obtaining a streetscape image emotion prediction model based on electroencephalogram signal recognition, and applying the streetscape image emotion prediction model to the output module to obtain a city streetscape visual emotion map; the method has the advantages of being sensitive in expression and accurate in emotion capture.

The invention relates to the field of smart cities, in particular to an urban landscape semantic segmentation method and system based on a large model, and the method comprises the steps: obtaining street view image data of a target geographic region; inputting the streetscape image data into a preset image semantic segmentation model, analyzing the streetscape image data through the image semantic segmentation model, and identifying various predefined environmental element categories contained in each image; calculating an urban landscape quantitative index based on the recognition result of the environmental element category; associating the calculated value of the urban landscape quantitative index to a corresponding road section or node in the road network spatial data corresponding to the target geographic area; generating a road network space visualization layer representing the value of the index value in a color grading manner based on the result after association; and according to a preset index threshold value, identifying the spatial road sections or nodes meeting the early warning conditions in the visual layer and generating prompt information. And the scientificity and comparability of an analysis result are ensured.

The invention discloses a land utilization classification method based on multi-source data intelligent fusion, and the method comprises the steps: obtaining multi-source geographic space data, including remote sensing image data, streetscape image data, building vector data, road network vector data, interest point data, mobile phone signaling data and microblog sign-in data, of a research region, and forming a land utilization classification experiment data set; remote sensing image features and streetscape image features of each research unit are constructed, environment features, interest point features and population distribution features are established, and a feature set of land utilization classification tasks is formed; combining the constructed features in pairs to generate feature pairs formed by features of different modalities, performing feature alignment on the features of different modalities through a comparative learning method, and adding the aligned features into a feature set of a land utilization classification task; and land utilization classification is carried out based on an XGBoost integrated learning model, and model validity inspection is carried out by adopting five-fold cross validation. According to the invention, the accuracy of land utilization classification is improved.

The invention discloses a dynamic street scene reconstruction method and device based on a four-dimensional Gaussian scene graph, which are used for synthesizing a new visual angle image of a dynamic street scene by optimizing four-dimensional Gaussian and realizing high-fidelity reconstruction of the street scene. The method comprises the following steps: giving an image of a dynamic street scene, a camera pose and image annotation information, respectively optimizing a static street background and a dynamic object through four-dimensional Gaussian, and combining into a four-dimensional Gaussian scene graph to represent the dynamic street scene; each node in the four-dimensional Gaussian scene graph stores a group of four-dimensional Gaussian parameters; and high-fidelity reconstruction of a dynamic street scene is realized based on the four-dimensional Gaussian of the graph. And giving any view angle of the dynamic street scene, selecting a corresponding four-dimensional Gaussian node from the four-dimensional Gaussian scene graph, and combining four-dimensional Gaussian according to the image annotation information to render the image of the dynamic street. According to the method, high-fidelity modeling of the dynamic street scene is realized through the four-dimensional Gaussian scene graph, and the rendering quality of the dynamic street scene is improved.

The invention provides a projection transformation method for hexahedral panoramic data to spherical panoramic data, and the method achieves the seamless splicing of panoramic images of six surfaces of a cube and the mapping of the panoramic images to a spherical coordinate system through the determination of a direction vector, the normalization of the vector, the conversion of spherical coordinates, the projection to an equidistant histogram, and an illumination and color correction algorithm. The problems of splicing gaps, deformation, texture distortion and the like in a traditional method are solved. The method is not only widely applied to the fields of virtual reality, augmented reality, panoramic video production and the like, but also greatly expands the application scenes, including but not limited to panoramic roaming, streetscape roaming experience, integrated use of specific roaming plug-ins, seamless embedding on a map platform and the like, and has wide application prospects. Therefore, the display effect of the panoramic data and the immersive experience of the user are remarkably improved.

The invention relates to the technical field of image segmentation, and discloses a street view image semantic segmentation system for a complex city scene, and the system comprises a double-edge energy ratio calculation module which obtains a double-edge energy ratio through calculation; the priori mask generation module is used for generating a priori mask; the segmentation network construction module is used for outputting logarithmic probabilities and probabilities of various categories through a trunk segmentation network; the total loss construction module is used for constructing total loss; the mutual suppression optimization module is used for calculating to obtain a preliminary semantic segmentation result; and the semantic segmentation output module is used for calculating to obtain a final semantic segmentation result. According to the method, a unified process of double-edge energy ratio driving direction convolution, separation loss and reasoning period mutual suppression is adopted, repeated boundaries and false objects caused by reflection and transmission mixing in a complex city scene containing glass and a water surface are reduced, pixel-level boundary positions are stabilized, and road space, building facades and traffic element division are consistent.

The invention relates to the field of city analysis, and discloses a method for analyzing the nonlinear influence of a city environment on city vitality, and the method comprises the steps: carrying out the preprocessing of multi-source geographic big data and multi-source remote sensing data; according to a Deeplab V3 + deep learning semantic segmentation algorithm, predicting the pixel ratio of each category in each streetscape image, and constructing streetscape features based on the pixel ratio; constructing urban vitality evaluation indexes, and aggregating the urban vitality evaluation indexes by using a TOPSIS algorithm to obtain an evaluation result of the urban vitality; constructing an urban environment index according to the preprocessed multi-source remote sensing data; analyzing the spatial distribution difference of the non-linear influence of the urban environment on the urban vitality by using a geographically weighted random forest algorithm; and analyzing the threshold effect of the non-linear influence of the urban environment on the urban vitality by using a Gaussian fitting line algorithm. According to the method, street view features are extracted through a semantic segmentation technology, and comprehensive, scientific and refined quantitative evaluation of urban vitality is realized.

The invention provides a panoramic streetscape image generation method and system, and belongs to the technical field of image data processing. According to the method, the complete panoramic image with real distortion is directly generated, a multi-view image splicing mode is not needed, and the high-fidelity panoramic streetscape image with global brightness consistency, seamless connection and real distortion can be generated on the basis of reserving accurate geometric control; moreover, the authenticity and consistency of the image are improved, the complexity of multi-view generation is reduced, and the applicability and real-time performance in an automatic driving scene are remarkably improved. Multi-scale geometric control and conditional coding are introduced, a pre-training diffusion model is combined, a panoramic streetscape image with real distortion is generated from multi-condition input of a road BEV map, a 3D target frame, a camera posture, text description and the like, geometric details such as road elevation and target object height can be accurately controlled in the generation process, and the image quality is improved. And the training effect of the 3D perception task is obviously improved.

The invention relates to an unsupervised field adaptive streetscape semantic segmentation method. The method comprises the following steps: establishing a PixEL unsupervised field adaptive network model; the method comprises the following steps: firstly, training a segmentation network Deeplabv2multi through a source domain data set to obtain a pre-trained segmentation model; perturbation is added to the features, obtained through a Deeplabv2multi encoder, of the image in a target domain, and a feature perturbation enhanced version of the image is obtained; performing consistency training on the strong enhancement version, the weak enhancement version and the feature disturbance version of the target domain image by using the segmentation model; in a target domain, energy scores of unmarked samples are calculated to evaluate the closeness degree of the unmarked samples and current training distribution, samples with higher energy scores are selected to generate pseudo labels, and the pseudo labels are utilized to further carry out optimization training on the model. Compared with an existing unsupervised field self-adaption method, the method has the advantages that a better semantic segmentation effect is achieved on a real reference data set through synthesis, the subjective vision and objective evaluation indexes are obviously improved, and the unsupervised field self-adaption problem in the street scene image semantic segmentation field is effectively solved.

The invention discloses a green vision index estimation model construction and green vision index monitoring method and device, and belongs to the technical field of geographic information, and the method comprises the steps: obtaining a plurality of remote sensing slice images and a plurality of streetscape images corresponding to the remote sensing slice images; wherein each remote sensing slice image corresponds to a road network; determining a sample label corresponding to the remote sensing slice image according to the plurality of streetscape images; calculating a corresponding auxiliary feature map according to the remote sensing slice image; wherein the auxiliary feature map comprises a normalized vegetation index map, a depth map and a visual field perception intensity map; and according to each remote sensing slice image, each auxiliary feature map and each sample label, training and verifying a preset deep learning model, and obtaining a green vision index estimation model. Therefore, by implementing the method and the device, the problem of how to perform large-range and periodic monitoring on the street green vision index under the pedestrian view angle can be solved.

The invention discloses a multi-fractal-based urban green vision rate dynamic evolution mode identification method. The method comprises the following steps: S100, acquiring urban street green vision rate data; s200, acquiring urban multi-scale walking isochronous circle data, and acquiring walking reachable areas within 5 minutes, 10 minutes, 15 minutes and 20 minutes through an API (Application Program Interface) of an OpenRouteService platform on the basis of an urban road network; s300, identifying a dynamic evolution mode of the green vision rate; and S400, identification result demonstration, including spatial distribution of a green vision rate dynamic evolution mode, key threshold identification and spatial structure differentiation, is used for assisting urban greening structure optimization and scientific intervention strategy formulation. According to the method, the problems of staticization, single scale and insufficient structure expression in the existing urban street view green visual rate analysis are solved.

The invention relates to the field of city planning, and discloses a city spatial evolution prediction system fusing multi-modal data, and the system comprises a data preprocessing and multi-modal knowledge graph construction module which is used for configuring a knowledge graph ontology architecture and preprocessing multi-source city data; establishing a multi-modal knowledge graph based on the preprocessed multi-source city data and the knowledge graph ontology architecture; the model setting module is used for configuring a multi-modal representation learning model; the data integration module is used for a multi-modal representation learning model and performing deep integration on multi-modal data in a vector space to obtain vector representation with remote sensing and streetscape semantic information; and the knowledge reasoning and predicting module is used for performing knowledge reasoning on the missing part in the multi-modal knowledge graph based on the vector representation so as to predict the urban spatial evolution event. The urban spatial evolution event prediction can be realized, and the problem that the remote sensing image and the streetscape image are difficult to be deeply fused is effectively solved.

The invention discloses a street space pleasure evaluation and optimization method based on multi-source data. The method comprises the following steps: data acquisition: acquiring street scene images, three-dimensional building data and POI data in a research area; subjective perception data acquisition: acquiring a subjective perception score of a user for street pleasure by using a street view image and an online data collection platform; objective index extraction: extracting objective indexes such as physical characteristics, interface characteristics, urban functions and macroscopic forms of streets from the multi-source data; correlation analysis: carrying out correlation analysis on objective indexes and subjective perception scores, and determining key factors influencing street pleasure; establishing an evaluation model: establishing a street space pleasure evaluation model based on the key factors; and optimization strategy making: proposing a targeted street design optimization strategy according to an evaluation result. According to the method, multi-source data and individual perception are comprehensively considered, the street space pleasure can be evaluated more accurately, and a scientific basis is provided for street design and planning.

The invention discloses a method for identifying a to-be-optimized space of a built environment based on a form hierarchical model. The method comprises the steps of collecting and preprocessing data of the built environment, preliminarily performing form hierarchical operation of the built environment, correcting the form hierarchical operation of the built environment, preliminarily identifying the to-be-optimized space, identifying the hierarchical operation of the to-be-optimized space, and interacting and feeding back identification results. According to the method, data acquisition is carried out through a laser radar system, an aerial photography unmanned aerial vehicle, a street scene acquisition vehicle and a near-earth satellite, and recognition and display of a to-be-optimized space of a built environment are carried out in combination with multi-task Bayesian federal learning, an XGBoost model, a Vision Transformer and a convolutional neural network. According to the method, the rapid identification of the to-be-optimized space of the built environment in the urban planning field can be assisted, and the rationality and accuracy of the identification result of the to-be-optimized space of the built environment are improved through form classification.

The invention discloses a multi-source fusion point cloud data driven three-dimensional streetscape generation and optimization method, which relates to the technical field of three-dimensional streetscape generation, and comprises the following steps of: according to a space range of input large-range point cloud data, combining a three-dimensional coordinate system distribution characteristic of the input large-range point cloud data; dividing the whole registration area into a plurality of uniformly distributed subunit areas with the same area by adopting a structured space division algorithm; for each subunit region, geometric structure features of internal point clouds of the subunit region are obtained, and a geometric feature descriptor containing multiple dimensions is constructed; according to the method, subunit structure complexity identification is realized through a heterogeneity evaluation mechanism based on geometric features and a support vector machine model, the upper limit of the number of matching points is dynamically regulated and controlled, the registration precision of a high-heterogeneous region is improved, and the mismatching risk of a low-heterogeneous region is reduced, so that the structure restoration and continuity of a streetscape model are optimized, and the robustness of the streetscape model is improved. And the stability and intelligence of the three-dimensional modeling system are enhanced.

The invention discloses a portable urban thermal environment intelligent sensing method and equipment based on AR, and the method comprises the steps: collecting a panoramic RGB street scene image of a user visual angle through AR glasses, and synchronously obtaining the time, geographic position and meteorological parameters; mapping the RGB image and the environmental parameters into a panoramic infrared thermal image by using a generative adversarial network, and obtaining the radiation temperature of each pixel; projecting the panoramic heat map to a unit spherical surface taking a user as a center, calculating a long-wave radiation flux through spherical surface angle integration, and calculating a short-wave radiation flux based on a solar radiation model by combining a solar elevation angle and the like; and fusing the total radiation amount to calculate the average radiation temperature, and finally superposing the heat map and / or the average radiation temperature to a real or virtual environment through AR glasses or an intelligent terminal to realize spatialization thermal load visualization.

Embodiments of the present application relate to the technical field of artificial intelligence. Disclosed is a training method for a street scene understanding model assisted by a large vision model, comprising: extracting a street scene instantaneous feature of the current street scene image frame, and extracting street scene robust invariant features of all historical street scene image frames; respectively acquiring the difference between the current street scene image frame and each historical street scene image frame, and extracting street scene dynamic features of all street scene difference images; randomly mixing the street scene instantaneous feature, the street scene robust invariant features, and the street scene dynamic features to obtain a mixed street scene feature; performing street scene semantic segmentation on the mixed street scene feature by means of an initial street scene understanding model to obtain a predicted street scene mask of the current street scene image frame; generating a sample street scene mask of the mixed street scene feature by means of a large vision model; and updating model parameters of the initial street scene understanding model on the basis of the predicted street scene mask and the sample street scene mask to obtain a target street scene understanding model. Thus, the accuracy of street scene understanding is improved.