Method and system for assessing urban cycling traffic resilience based on street view perception and machine learning

Abstract

This invention discloses a method and system for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning, relating to the field of traffic scene image processing technology. The method includes: acquiring a street view image sequence of the target cycling segment; extracting the canopy region, bike lane region, sky region, and shadow region using an image semantic segmentation model; generating an observation area above the center track based on the bike lane region; calculating the lateral misalignment between the canopy region and the bike lane region, the sky opening of the center track, and the degree to which the bike lane is not covered by shadow; generating a reverse thermal resilience risk value based on the proportion of the canopy region; and calculating a continuous sun exposure risk value based on the risk values ​​at each sampling location and their continuity, thus generating a cycling traffic thermal resilience assessment result. This invention can identify the risk of continuous sun exposure on the center track of a bike lane under high green visibility conditions, improving the accuracy of the assessment.

Description

Technical Field

[0001] This invention relates to the field of traffic scene image processing technology, and in particular to a method and system for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning. Background Technology

[0002] With the increasing frequency of extreme heat events, the suitability of urban cycling for commuting, slow-traffic system construction, and low-carbon transportation organization is significantly affected by the road thermal environment. For cyclists, actual heat exposure depends not only on the overall urban temperature but also on the distribution of tree canopies along the roadside, the location of the bike lane, the size of the sky gap, and the amount of shade covering the bike lane surface. In some urban roads, street view images show a large number of tree canopies and a high overall green visibility, but the canopies are mainly distributed on the sidewalk side, the road edge side, or outside the separation strip between motorized and non-motorized vehicles, resulting in a lateral misalignment with the center line of the bike lane. This leads to a significant sky gap above the bike lane, and the bike lane surface lacks effective shade coverage. In such cases, cyclists may still be continuously exposed to the sun while traveling along the bike lane, creating a reverse thermal resilience risk where the streetscape appears well-greened but the cyclist's thermal protection is insufficient.

[0003] Existing methods for evaluating urban cycling environments or road thermal environments often rely on streetscape green visibility, canopy coverage, road greening levels, or regional average thermal environment indicators. Roads with high visible green coverage are typically considered to have good shading conditions and high-temperature adaptability. However, these methods often fail to further analyze the correspondence between the canopy area and the bike path area in the horizontal direction of the image, and do not jointly assess the sky opening above the bike path's center track and the shadow coverage on the bike path surface. This easily overlooks the spatial misalignment between the visible green coverage of the canopy and the effective shading line of the bike path. Under extreme high-temperature conditions, existing evaluation methods can easily misclassify roads with high streetscape green visibility but continuous sun exposure along the bike path's center track as having high thermal resilience, leading to inconsistencies between the urban cycling traffic thermal resilience assessment results and the actual sun exposure risk for cyclists. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as the disconnect between streetscape greening evaluation and actual shading evaluation of bicycle lanes, making it difficult to accurately identify the risk of continuous sun exposure for cycling routes under the appearance of high green visibility. The invention proposes a method and system for assessing the thermal resilience of urban cycling traffic based on streetscape perception and machine learning.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution:

[0006] Urban cycling traffic thermal resilience assessment methods based on street view perception and machine learning include:

[0007] A street view image sequence is acquired for thermal toughness assessment of the target cycling route. A trained image semantic segmentation model is used to segment each street view image in the street view image sequence to extract the canopy region, bike lane region, sky region, and shadow region.

[0008] Extract the centerline of the bicycle lane area, and generate an observation area above the center trajectory in the street view image based on the centerline;

[0009] Calculate the projections of the tree canopy region and the bicycle lane region onto the horizontal direction of the street view image, respectively. Based on the overlap relationship between the projections of the tree canopy region and the bicycle lane region, calculate the lateral misalignment between the tree canopy region and the bicycle lane region.

[0010] The sky opening of the central trajectory is calculated based on the intersection of the observation area above the central trajectory and the sky area.

[0011] The reverse thermal toughness risk value of the street view image is calculated based on the proportion of the canopy area in the street view image, the coverage of the shadow area in the bicycle lane area, the lateral misalignment, and the sky opening of the center trajectory.

[0012] Based on the reverse thermal toughness risk value of each street view image and the arrangement order of each street view image in the target cycling section, a continuous exposure risk value is calculated, and a thermal toughness assessment result of the target cycling section is generated based on the continuous exposure risk value.

[0013] Preferably, the canopy area, bike path area, sky area, and shade area are extracted, including:

[0014] A trained image semantic segmentation model is used to perform pixel-level segmentation on each street scene image to obtain pixel sets belonging to tree canopy, bike path surface, sky opening and shadow respectively;

[0015] The set of pixels belonging to the tree canopy is defined as the tree canopy region;

[0016] The set of pixels belonging to the surface of the bicycle lane is defined as the bicycle lane region;

[0017] The set of pixels belonging to the sky opening is defined as the sky region;

[0018] The set of pixels that belong to the shadow is defined as the shadow region.

[0019] Preferably, the centerline of the bicycle lane area is extracted, and an observation area above the center trajectory is generated above the street view image based on the centerline, including:

[0020] The bicycle lane area is skeletonized to obtain the centerline representing the bicycle lane traffic position.

[0021] For any pixel on the center line, the pixels located at the horizontal position of the pixel along the vertical direction of the street view image to the upper boundary of the street view image are included in the candidate region.

[0022] All candidate regions corresponding to the center line are merged to obtain the observation area above the center trajectory.

[0023] Preferably, the projections of the tree canopy region and the bicycle lane region onto the horizontal direction of the street view image are calculated respectively. Based on the overlap relationship between the projections of the tree canopy region and the bicycle lane region, the lateral misalignment between the tree canopy region and the bicycle lane region is calculated, including:

[0024] Calculate the normalized projections of the tree canopy region and the bicycle lane region along the lateral direction of the street view image, respectively;

[0025] For each lateral coordinate of the street view image, determine the minimum value between the normalized projection of the canopy region and the normalized projection of the bicycle lane region;

[0026] The minimum values ​​corresponding to all horizontal coordinates are summed to obtain the horizontal overlap.

[0027] Subtract the lateral overlap from the value 1 to obtain the lateral misalignment.

[0028] Preferably, the sky aperture of the central trajectory is calculated based on the intersection of the observation area above the central trajectory and the sky area, including:

[0029] Obtain the total number of pixels within the observation area above the central trajectory;

[0030] Obtain the number of pixels in the overlapping area between the observation region above the central trajectory and the sky region;

[0031] Calculate the ratio of the number of pixels in the overlapping intersection region to the total number of pixels in the observation area above the central trajectory, and use the ratio as the sky aperture of the central trajectory.

[0032] Preferably, the reverse thermal toughness risk value of the street view image is calculated based on the proportion of the canopy area in the street view image, the coverage of the shadow area within the bicycle lane area, the lateral misalignment, and the sky opening of the center trajectory, including:

[0033] The ratio of the number of pixels in the tree canopy region to the total number of effective pixels in the street view image is calculated to obtain the proportion of the tree canopy region.

[0034] The number of overlapping pixels between the bike lane area and the shadow area is obtained, and the ratio of the number of overlapping pixels to the total number of pixels in the bike lane area is calculated to obtain the coverage of the shadow area in the bike lane area.

[0035] Subtract the coverage of the shaded area within the bike lane area from the value 1 to obtain the degree to which the bike lane is not covered by shadow;

[0036] The reverse thermal toughness risk value is obtained by multiplying the canopy area ratio, the lateral misalignment, the sky opening of the central trajectory, and the degree to which the bicycle path is not covered by shadow.

[0037] Preferably, the continuous exposure risk value is calculated based on the reverse thermal toughness risk value of each of the street view images and the arrangement order of the street view images in the target cycling segment, including:

[0038] According to the order in which the sampling positions of each street view image are arranged in the target cycling segment, the reverse thermal toughness risk values ​​of each street view image are combined into a risk sequence.

[0039] Calculate the cumulative sum of all reverse thermal toughness risk values ​​in the risk sequence;

[0040] Calculate the square root of the product of two adjacent reverse thermal toughness risk values ​​in the risk sequence, and sum the square roots of all adjacent two reverse thermal toughness risk values ​​to obtain the sum of adjacent continuous risks.

[0041] The cumulative sum is added to the sum of adjacent consecutive risks, and then divided by the normalization coefficient to obtain the continuous sun exposure risk value.

[0042] The normalization coefficient is twice the total number of images in the street view image sequence minus one.

[0043] Preferably, generating a thermal resilience assessment result for the target cycling section based on the continuous exposure risk value includes:

[0044] The continuous exposure risk value is used as the reverse evaluation value of the thermal toughness of the target cycling section;

[0045] The risk value of continuous sun exposure is negatively correlated with the thermal toughness of cycling traffic on the target cycling route.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1. This invention acquires street view image sequences of target cycling routes and uses a trained image semantic segmentation model to extract canopy areas, bike lane areas, sky areas, and shadow areas. This allows the assessment of cycling thermal resilience to move beyond the overall green visibility of the street view or the visibility ratio of the canopy, and further focus on the actual traffic lane positions. By extracting the centerline from the bike lane area and generating an observation area above the center trajectory, the determination of sky openings can be limited to above the actual center trajectory traversed by the cyclist. By calculating the lateral misalignment between the canopy area and the bike lane area in the horizontal direction of the image, it can identify spatial misalignment states where the canopy is visible but does not cover the bike lane position. It can effectively identify special road sections where the street view greening appears good but the center trajectory of the bike lane is still at risk of sun exposure, thus improving the pertinence and accuracy of urban cycling traffic thermal resilience assessment.

[0048] 2. This invention further incorporates the proportion of the canopy area, lateral misalignment, sky opening at the center trajectory, and the degree of lack of shade coverage on the bike path into the calculation of the reverse thermal resilience risk value. This allows the assessment results to simultaneously reflect the combined relationship between factors such as visible green canopy, bike path misalignment, sky opening above, and lack of surface shade coverage, avoiding misjudgments caused by a single indicator. Following the arrangement of street view images in the target cycling route, the reverse thermal resilience risk values ​​at each sampling location are constructed into a risk sequence. Combined with continuous risk items between adjacent sampling locations, a continuous sun exposure risk value is calculated, thereby identifying the sun exposure risk under a continuous high green visibility appearance along the cycling direction. This not only outputs the overall reverse thermal resilience evaluation results for the target cycling route but also helps locate specific sections with insufficient shading along the cycling path, providing more interpretable image analysis evidence for optimizing the urban slow-traffic environment, adjusting canopy configuration, and improving the thermal resilience of cycling paths under extreme high-temperature conditions. Attached Figure Description

[0049] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0050] Figure 1 This is a flowchart illustrating an embodiment of the urban cycling traffic thermal resilience assessment method based on street view perception and machine learning provided by the present invention.

[0051] Figure 2 This is a functional block diagram of an urban cycling traffic thermal resilience assessment system based on street view perception and machine learning, provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Example: This example provides a method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning. See [link to example]. Figure 1 Specifically, it includes the following steps:

[0054] S1. Obtain the street view image sequence and perform image semantic segmentation;

[0055] Specifically, obtain street view image sequences for thermal toughness assessment of the target cycling route:

[0056]

[0057] in, This represents the street view image corresponding to the i-th sampling location in the target cycling route. , where n represents the total number of images in the street view image sequence; street view images The horizontal coordinates of the image are denoted as x, and the vertical coordinates are denoted as y. In the image coordinates, the horizontal direction represents the pixel position of the road cross-section in the street view image in the left and right direction, and the vertical direction represents the direction extending from the surface of the bicycle lane to the upper boundary of the image. The vertical direction is not limited to the increase or decrease of the vertical coordinate value, but is based on the direction of the upper boundary of the street view image. The street view images are arranged in the order of sampling positions in the target cycling segment so that the reverse thermal toughness risk value calculated later can reflect the continuous distribution of risk along the cycling direction.

[0058] In this embodiment, thermal resilience assessment refers to the image-based evaluation of the spatial shading protection capability of the target cycling route in response to the risk of sun exposure that cyclists may encounter under high temperature weather. The thermal resilience assessment does not limit the street view images to be collected on the day of extreme high temperature. Instead, it uses the canopy area, bike lane area, sky area and shade area in the street view images that can reflect the road spatial structure to determine whether the central passage of the bike lane is likely to be under continuous sun exposure under extreme high temperature conditions.

[0059] In this embodiment, the target cycling route is an urban road segment with a bicycle lane or non-motorized vehicle cycling area; the street view image sequence can be obtained by vehicle-mounted street view acquisition equipment, road inspection equipment, mobile terminal, or an existing street view database; the street view image sequence is used for thermal resilience assessment of the target cycling route, and the images themselves can be images acquired during the thermal resilience assessment period, or historical street view images that can reflect the road spatial structure, canopy distribution, bicycle lane location, sky opening and shadow status of the target cycling route; each street view image is sorted according to the cycling direction or road travel direction of the target cycling route;

[0060] The thermal resilience assessment in this embodiment refers to evaluating the road space conditions of the target cycling route under extreme high-temperature conditions, which are prone to causing sun exposure risks for cyclists. The street view image sequence is mainly used to identify the visible position of tree canopies, the position of the bike lane, the position of the sky opening, and the distribution of shadows in the target cycling route, thereby determining whether the visible tree canopies in the street view actually affect the center line of the bike lane. The street view images can be images collected during periods of extreme high temperatures or historical street view images that can reflect the road space structure and shading relationship of the target cycling route. When the street view image is collected during a period of extreme high temperatures, the shadow area can reflect the actual shading state of the bike lane surface during that period. When the street view image is a historical street view image, the tree canopy area, the bike lane area, and the sky area are used to reflect the relative positional relationship between the visible green amount of tree canopies and the center line of the bike lane in the road space, and the shadow area is used to help characterize whether there is a visible shading state on the bike lane surface. This embodiment does not only use the green visibility rate of the street view as the basis for thermal resilience assessment, but also uses the spatial correspondence in the street view image to identify the sun exposure risk of the cycling route under the appearance of high green visibility rate.

[0061] Each street view image is segmented using a trained image semantic segmentation model. Pixel-level segmentation is performed; the image semantic segmentation model is trained using street scene images labeled with tree canopies, bike path surfaces, sky openings, and shadows as training samples, enabling it to classify pixels in the input street scene image and obtain four types of regions directly related to the subsequent calculations in this embodiment; the image semantic segmentation model can employ a deep learning image segmentation network with pixel-level classification capabilities, and the specific network structure of the model is not a limitation of this embodiment, as long as it can output the semantic category corresponding to each pixel;

[0062] For the i-th street view image After performing image semantic segmentation, the tree canopy region is obtained. Bicycle lane area Sky area and shaded areas Among them, the canopy area The set of pixels in a street view image that belong to the canopy of roadside trees, greenery, or vegetation covering the road; bicycle lane area. The set of pixels in a street view image that belong to the surface of a bike lane or a non-motorized vehicle riding area; the sky area. The set of pixels in the street view image that belong to the sky opening; shadow area The set of pixels belonging to the shadow area in the street view image is used to calculate the intersection with the bicycle lane area to determine the shadow coverage of the bicycle lane surface.

[0063] In this embodiment, the canopy area Bicycle lane area Sky area and shaded areas All are from the same street view image The pixel set obtained by segmentation is located in the same image coordinate system; the canopy region The bicycle path area is used to represent the spatial distribution of visible tree canopies in an image. The sky area is used to indicate the actual location of the road surface traveled by the cyclist. The shaded area is used to indicate an opening in the sky above a bike path where direct sunlight may occur. Used to indicate the visible shaded area that can overlap with the bicycle lane area; in this embodiment... , , and Both can be represented as street view images Binary masks of the same size; using the canopy region For example, when pixel When it is within the canopy area, When pixel When not within the canopy area, Bicycle lane area Sky area and shaded areas The binary representation is the same; through the above binary mask, the number of pixels, region intersection and lateral projection of each region can be calculated.

[0064] S2. Generate the observation area above the center trajectory based on the bicycle lane area;

[0065] For the bike lane area in the i-th street view image Centerline extraction is performed; specifically, for the bicycle lane area. Skeletonization is performed to obtain the centerline representing the bike lane's traffic position; the centerline is used to characterize the central trajectory position that a cyclist is most likely to pass through when traveling along the bike lane, rather than the center position of the entire street view image or the center position of the entire road.

[0066] In this embodiment, the center line refers to the line defined by the bicycle lane area. The linear pixel set obtained from the image morphology extraction is used to represent the representative travel line position of a cyclist traveling along the bike lane; the centerline does not represent the overall road centerline, nor the centerline of the motor vehicle lane, but rather the centerline of the bike lane area. Direct correspondence;

[0067] In one specific implementation, the bicycle lane area is first... The binary mask is used to determine the connected regions, and the effective connected regions corresponding to the surface of the bicycle lane are retained. Then, the effective connected regions are refined so that the bicycle lane region is gradually shrunk into a skeleton line with a single pixel or approximately single pixel width that maintains its geometric extension direction, thereby obtaining the center line of the bicycle lane region. This center line is used as the line position reference for generating the observation area above the center trajectory in this step.

[0068] For any pixel on the center line, let this pixel be denoted as p, and its image coordinates be... At the vertical position of this pixel, based on the bicycle lane area The left and right boundaries of the image determine the local width of the bike lane at that location. A horizontal neighborhood corresponding to the local width of the bike lane is selected in the horizontal direction of the image, centered on pixel p. Pixels within this horizontal neighborhood, extending from the top of the street view image's vertical axis to the top boundary of the street view image, are included in the candidate region. Candidate regions are generated for all pixels along the center line, and all candidate regions are merged to obtain the observation area above the center trajectory. ;

[0069] The local width of the bicycle lane is defined by the bicycle lane area at the vertical position of the pixel. The pixel distance between the left and right boundaries is determined; if the left and right boundaries of the bike lane area cannot be obtained simultaneously at this vertical position, the local width of the bike lane that can be obtained at the position adjacent to the center line is used to generate the horizontal neighborhood; the observation area above the center trajectory. It not only covers the space above the single column of pixels of the center line, but also covers the field of view above the actual passage position corresponding to the cyclist when traveling along the center line of the bike lane, thereby reducing the impact of single column pixel noise on the calculation of the sky opening of the center trajectory.

[0070] In this embodiment, the observation area above the central trajectory This refers to the set of pixels extending vertically upwards in a street view image, using the centerline of the bike lane as a reference. It is used to detect whether there is an opening in the sky above a cyclist traveling along the centerline of the bike lane; the observation area above the center track... Its function is to limit the determination of the sky opening to the field of view above the center line of the bicycle lane;

[0071] By building In this embodiment, the subsequent determination of the sky opening is limited to above the center track of the bicycle lane, rather than making an average determination over the entire street view image or the entire road. This can more accurately reflect the exposure status of the actual route of the cyclist.

[0072] S3. Calculate the lateral misalignment between the canopy area and the bicycle path area;

[0073] Specifically, based on the obtained canopy area and bicycle lane area The normalized projections of the two regions in the horizontal direction of the street view image are calculated respectively. The horizontal projection is used to characterize the distribution of the canopy area and the bike lane area in the horizontal position of the image, so as to determine whether the visible canopy is located near the horizontal position of the bike lane.

[0074] In this embodiment, the lateral direction refers to the image coordinate direction corresponding to the left-right extension relationship of the road cross-section in the street view image; the lateral projection refers to the projection of the tree canopy area... or bicycle lane area The pixel distribution is accumulated along the vertical axis of the image to obtain its distribution on the horizontal axis of the image; by horizontal projection, it is possible to determine whether the visible tree canopy and bicycle lane area in the street view image correspond in the horizontal position without restoring the real three-dimensional distance of the road.

[0075] Canopy area The normalized projection along the horizontal direction of the image is calculated as follows:

[0076]

[0077] in, This represents the normalized projection value of the tree canopy region at the horizontal coordinate x; Represents pixels Does it belong to the canopy area? This represents the number of pixels belonging to the tree canopy region in the vertical pixel column corresponding to the horizontal coordinate x; This represents the total number of pixels in the tree canopy region. Through normalization, even if the number of tree canopy pixels differs between different images, the distribution of the tree canopy region in the horizontal direction can be compared.

[0078] Bike lane area The normalized projection along the horizontal direction of the image is calculated as follows:

[0079]

[0080] in, This represents the normalized projection value of the bicycle lane area at the horizontal coordinate x; Represents pixels Is it within a bicycle lane area? This indicates the number of pixels belonging to the bicycle lane area in the vertical pixel column corresponding to the horizontal coordinate x; This represents the total number of pixels in the bike lane area.

[0081] In obtaining and Then, for each horizontal coordinate x in the street view image, the minimum value between the normalized projection of the canopy region and the normalized projection of the bike lane region is determined, and the minimum values ​​corresponding to all horizontal coordinates are summed to obtain the horizontal overlap. The larger the horizontal overlap, the closer the distribution of the canopy region and the bike lane region is in the horizontal direction of the image; the smaller the horizontal overlap, the more the canopy region and the bike lane region are in the horizontal position.

[0082] Based on this, subtract the lateral overlap from the value of 1 to obtain the lateral misalignment between the canopy area and the bike path area. :

[0083]

[0084] in, This represents the lateral misalignment at the sampling position corresponding to the i-th street view image; This represents the smaller of the normalized projections of the canopy area and the bicycle path area at the horizontal coordinate x. This indicates the degree of overlap between the tree canopy area and the bicycle path area in terms of their lateral distribution;

[0085] In this embodiment, the lateral misalignment Refers to the canopy area Bicycle lane area Lateral distribution deviation under the same street view image coordinate system; lateral misalignment The larger the value, the further the visible tree canopy in the street view image is from the lateral position of the bike lane, and the lower the likelihood that the visible greenery of the tree canopy will translate into effective shade for the bike lane; lateral misalignment The smaller the value, the closer the canopy area and the bike path area are in the horizontal position of the image, and the more likely the visible green of the canopy will affect the bike path.

[0086] Lateral misalignment in this embodiment Used to characterize the relative lateral correspondence between the visible area of ​​tree canopy and the visible area of ​​bike lane in the same street view image coordinate system; due to the tree canopy area and bicycle lane area All images are from the same street view image. The pixel-level segmentation results, both having the same image coordinate reference, can be compared in terms of their relative distribution in the horizontal direction of the image through horizontal normalized projection; the horizontal misalignment degree It does not require restoring the real-world distance between the tree canopy area and the bike lane area, but is used to determine whether the visible tree canopy in the street view image corresponds to the location of the bike lane in the horizontal position of the image. If it is necessary to further reduce the influence of the street view perspective relationship on the horizontal projection, perspective correction can be performed on the street view image or the bike lane area before calculating the horizontal normalized projection, so that the bike lane area is closer to the road plane projection relationship in the image.

[0087] when When the value is large, it indicates that although there may be many tree canopies in the street view image, the canopies are mainly distributed laterally away from the bike lane, such as on the sidewalk side, outside the median strip, or above the road edge, and the possibility of the center line of the bike lane being obscured by tree canopies is low; when When the value is smaller, it indicates that the tree canopy area is closer to the bike path area in the lateral direction, and the bike path is more likely to be shaded by the tree canopy.

[0088] S4. Calculate the sky opening degree of the center trajectory;

[0089] Specifically, the sky opening degree of the center track is used to characterize whether there is a sky opening above the center line of the bike lane. The more obvious the sky opening, the more likely the cyclist is to be exposed to direct sunlight when passing along the center line.

[0090] In this embodiment, the sky aperture of the center trajectory is... The observation area above the center trajectory The area belongs to the sky region The pixel ratio; this metric reflects the degree to which the center line of the bike lane is not obscured by tree canopies, buildings, or other obstructions; the sky opening of the center track. The sky proportion is not equivalent to the entire street view image; the evaluation object is limited to the observation area above the center track of the bicycle lane.

[0091] Center trajectory sky opening The calculation method is as follows:

[0092]

[0093] in, This represents the sky aperture of the center trajectory at the sampling location corresponding to the i-th street view image; Indicates the sky area; Indicates the observation area above the central trajectory; This indicates the overlap and intersection of the sky region and the observation area above the central trajectory; This indicates the number of pixels in the overlapping intersection region; This represents the total number of pixels within the observation area above the center trajectory; when A larger value indicates that a significant proportion of pixels within the observation area above the center track belong to the sky, suggesting a clear sky opening above the bike path's center track, where tree canopies, buildings, or other obstructions fail to effectively cover the area above the cycling lane. A smaller value indicates fewer openings in the sky above the central trajectory, suggesting strong occlusion above the cycling line.

[0094] S5. Calculate the reverse thermal toughness risk value of the street view image;

[0095] In this embodiment, the reverse thermal toughness risk value This refers to an evaluation value used to characterize whether a single street scene sampling location has high tree canopy visibility but the center line of the bicycle path is still exposed to direct sunlight; the reverse thermal toughness risk value It's called "reverse" because it focuses on the anomaly in street view images where tree canopies have high visible green coverage but insufficient shading along cycling routes, rather than simply evaluating whether more tree canopies equate to greater thermal resilience; Reverse Thermal Resilience Risk Value The larger the value, the more likely the sampling location is to have a risk of sun exposure on the bike path despite the apparent high green visibility.

[0096] Specifically, the canopy area proportion is calculated; the canopy area proportion is used to characterize the visible green content of tree canopies in street view images, and its calculation method is as follows:

[0097]

[0098] in, Indicates the canopy area The number of pixels, Street view image The total number of effective pixels; the total number of effective pixels can be understood as the total number of pixels participating in image segmentation and subsequent calculations; in this embodiment, the visible green amount of the canopy refers to the proportion of pixels that can be identified as the canopy area in the street view image by the image semantic segmentation model or the visible distribution state of the canopy area, which is used to characterize the degree of presence of the canopy in the street view image at the visual level; the visible green amount of the canopy is different from the effective shading ability of the bicycle lane, and a higher visible green amount of the canopy only indicates that there are more visible canopies in the street view image;

[0099] Calculate the coverage of the shaded area within the bike lane area; the coverage of the shaded area within the bike lane area is used to characterize the degree to which the bike lane surface is actually covered by shadow, and it is calculated as follows:

[0100]

[0101] in, Indicates the shaded area Bicycle lane area The overlapping intersection; This indicates the number of pixels within the bike lane area that are in the shaded region. This represents the total number of pixels in the bike lane area; subtract the coverage of the shadow area within the bike lane area from the value of 1 to obtain the degree to which the bike lane is not covered by shadow.

[0102] In this embodiment, the degree to which the bicycle lane is not covered by shadow refers to the bicycle lane area. The middle and shaded areas The percentage of overlapping pixels is used to characterize the degree to which the surface of the bike path is not in shadow; the higher the degree to which the bike path is not covered by shadow, the less actual shading the bike path surface is; the lower the degree to which the bike path is not covered by shadow, the higher the degree to which the bike path surface is covered by shadow.

[0103] The method for calculating the degree to which a bike lane is not covered by shadow is as follows:

[0104]

[0105] The greater the degree of lack of shade on a bike path, the higher the proportion of the bike path surface that is not in shade, and the higher the likelihood that cyclists will be exposed to direct sunlight or high radiation when passing through that area; the smaller the degree of lack of shade on a bike path, the more adequate the shade coverage on the bike path surface.

[0106] Multiplying the canopy area ratio, lateral misalignment, sky opening at the center trajectory, and the degree to which the bike path is not covered by shadows yields the inverse thermal toughness risk value. :

[0107]

[0108] in, This represents the reverse thermal toughness risk value at the sampling location corresponding to the i-th street view image; Indicates the proportion of the tree canopy area; Indicates the lateral misalignment between the tree canopy area and the bicycle path area; Indicates the sky opening degree of the center trajectory; This indicates the degree to which the bike path is not covered by shadow; the calculation logic of this formula is: when the canopy area accounts for a high proportion in the street view image, it indicates that the street view image has a significant visible green area; when the lateral misalignment is... A higher value indicates that these visible tree canopies do not correspond to the bike path area in a lateral position; when the sky opening at the center track is high... A high level indicates that there is still a significant opening in the sky above the centerline of the bike path; a high level of non-shading indicates that the actual shadow coverage on the bike path surface is insufficient; only when all of the above factors are high simultaneously... This will significantly increase the ability to identify the risk of reverse thermal toughness in areas where streetscape greenery is high but the center track of the bike path is still under continuous sun exposure.

[0109] In the calculation process of this embodiment, the anomaly recognition results in the street view image can also be processed; if the street view image No bicycle lane area was identified in the middle If the street view image cannot represent the cycling lane position in the target cycling section, the street view image will be marked as an invalid sample and will not be included in the calculation of the continuous sun exposure risk value of the target cycling section.

[0110] In this embodiment, invalid samples refer to street view images that cannot provide the center line of the bicycle lane, cannot generate the observation area above the center track, or cannot participate in the calculation of the reverse thermal toughness risk value. Street view images marked as invalid samples are not included in the construction of risk sequences and the calculation of continuous exposure risk values, so as to avoid the distortion of evaluation results due to the lack of key image areas.

[0111] If street view image Canopy area not identified Then, the proportion of the tree canopy area corresponding to the street view image is set to zero, and the reverse thermal toughness risk value of the street view image is set to zero. The value is zero, thus avoiding the risk of misjudging images with no visible green canopy as having reverse thermal resilience under the illusion of high green visibility; if street view images There are bicycle lane areas However, the shadow area was not identified. If the shaded area has zero coverage within the bike lane area, then the bike lane not covered by the shade has a coverage of one. If based on the bike lane area... Failed to generate observation area above the center trajectory If so, the street view image is marked as an invalid sample, or it is re-based on the bike lane area. The observation area above the center trajectory generated by the boundary centerline Street view images marked as invalid samples are removed when forming the risk sequence, and the risk sequence is rearranged according to the sampling position order of the remaining valid street view images in the target cycling segment.

[0112] S6. Calculate the risk value of continuous exposure to sunlight and generate thermal toughness assessment results;

[0113] Specifically, when there are street view images marked as invalid samples in the street view image sequence, the invalid samples are first removed from the street view image sequence, and the effective risk sequence is determined according to the original sampling position order of the remaining effective street view images in the target cycling segment; to avoid repetition of variable meanings, this embodiment denotes the number of effective samples in the effective risk sequence as m, and the reverse thermal toughness risk value corresponding to the l-th effective sample as m. ,in, , This represents the sampling position number of the l-th valid sample in the original street view image sequence; if there are no invalid samples in the street view image sequence, then m = n, and ;

[0114] Based on the sampling location order of each valid street view image in the target cycling segment, the reverse thermal toughness risk values ​​of each valid street view image are combined into a risk sequence:

[0115]

[0116] in, represents the reverse thermal toughness risk value at the sampling location corresponding to the l-th valid sample, and m represents the number of valid samples;

[0117] Calculate the cumulative sum of all inverse thermal toughness risk values ​​in the risk sequence:

[0118]

[0119] This cumulative sum is used to characterize the overall level of reverse thermal toughness risk at each effective sampling location within the target cycling segment;

[0120] Further calculate the square root of the product of two adjacent reverse thermal toughness risk values ​​in the risk sequence, and sum the square roots of all adjacent pairs of reverse thermal toughness risk values ​​to obtain the sum of adjacent consecutive risks:

[0121]

[0122] in, and This represents the reverse thermal toughness risk value corresponding to two adjacent valid sampling locations in the target cycling route; This indicates the degree of risk continuity between two adjacent valid sampling locations; when two adjacent valid sampling locations... and When both are relatively high, The relatively large value indicates that the risk of sun exposure on bike lanes under the illusion of high green visibility exists continuously along the cycling direction; when the risk at one of the effective sampling locations is low, the adjacent continuous risk item does not increase significantly, indicating that the risk of sun exposure does not form a clear continuous segment;

[0123] The cumulative sum is added to the sum of adjacent continuous risks, and then divided by the normalization coefficient to obtain the continuous sun exposure risk value Q. In this embodiment, the continuous sun exposure risk value Q is an evaluation value used to characterize the degree of continuous existence of reverse thermal toughness risk along the cycling direction within the target cycling segment. The continuous sun exposure risk value Q considers both the reverse thermal toughness risk value of each street scene sampling location and whether the risk is continuous between adjacent sampling locations. When multiple adjacent sampling locations have high reverse thermal toughness risk values, the continuous sun exposure risk value Q increases, indicating that cyclists may experience continuous sun exposure risk when traveling along this segment. When high risk only occurs in isolation at local sampling locations, the continuous sun exposure risk value Q will not be significantly amplified.

[0124] The risk value Q from continuous sun exposure is calculated as follows:

[0125]

[0126] Where Q represents the risk value of continuous sun exposure for the target cycling route; This represents the cumulative sum of the reverse thermal toughness risk values ​​for all valid sampling locations; This represents the sum of adjacent consecutive risks; This represents the normalization coefficient. Since there are m single-point risk values ​​in the effective risk sequence and m-1 adjacent consecutive risk terms, the normalization coefficient is taken as 2m-1 to ensure that the continuous exposure risk value can simultaneously reflect the single-point risk level and the adjacent consecutive risk level.

[0127] The thermal toughness assessment results of the target cycling section are generated based on the continuous exposure risk value Q; specifically, the continuous exposure risk value Q is used as the reverse evaluation value of the thermal toughness of the target cycling section.

[0128] In this embodiment, the thermal toughness inverse evaluation value refers to the evaluation result that is negatively correlated with the thermal toughness of cycling traffic. The larger the thermal toughness inverse evaluation value, the more likely the target cycling route is to be continuously exposed to the sun under extreme high temperature conditions, and the weaker its cycling traffic thermal toughness. The smaller the thermal toughness inverse evaluation value, the lower the risk of misalignment between the visible green amount of the tree canopy and the effective shading line of the bicycle lane in the target cycling route, and the stronger its cycling traffic thermal toughness.

[0129] The continuous exposure risk value Q is negatively correlated with the thermal resilience of the target cycling route. That is, the larger the continuous exposure risk value Q, the more likely the target cycling route is to experience a reverse thermal resilience state where the visible green content of the tree canopy is high but the center track of the bicycle path is continuously exposed to the sun, and the weaker the thermal resilience of the target cycling route. The smaller the continuous exposure risk value Q, the lower the risk of lateral misalignment between the visible green content of the tree canopy and the effective shading line of the bicycle path in the target cycling route, the lower the degree of continuous exposure of the center track of the bicycle path, and the stronger the thermal resilience of the target cycling route.

[0130] The thermal resilience assessment results generated in this embodiment can include the continuous exposure risk value Q of the target cycling segment, the reverse thermal resilience evaluation results of the target cycling segment, and the reverse thermal resilience risk value sequence corresponding to each valid street view image. For a single target cycling segment, the continuous exposure risk value Q and its corresponding reverse thermal resilience evaluation results can be directly output. For multiple target cycling segments, they can be sorted from largest to smallest according to the continuous exposure risk value Q. Target cycling segments with larger continuous exposure risk values ​​Q are identified as segments with weaker thermal resilience under extreme high temperatures, while target cycling segments with smaller continuous exposure risk values ​​Q are identified as segments with stronger thermal resilience under extreme high temperatures. The thermal resilience assessment results can not only provide an overall evaluation of the target cycling segment, but also retain the reverse thermal resilience risk values ​​at each sampling location, which can be used to locate specific segments with high street view green visibility but continuous exposure to sunlight in the center of the bike path.

[0131] This embodiment enables the identification of risk conditions from street view images that are difficult to detect with ordinary green visibility assessments under extreme high-temperature assessment scenarios. The road has a high visual green canopy coverage, but the canopy position is laterally misaligned with the center line of the bicycle lane. There are still gaps in the sky above the bicycle lane and the surface of the bicycle lane lacks shade coverage. This avoids directly equating the overall greening level of the street view with the shading capacity of the cycling lane, and improves the accuracy and interpretability of urban cycling traffic thermal resilience assessment.

[0132] Example 2, based on Example 1, further illustrates the training and usage of the image semantic segmentation model;

[0133] Specifically, a street view image training sample set is constructed. The street view image training sample set includes multiple urban road street view image samples, and each street view image sample corresponds to a pixel label image with the same size as the street view image sample. In the pixel label image, each pixel is labeled as one of the following categories: tree crown category, bicycle lane surface category, sky opening category, shadow category, or background category. The tree crown category is used to label the tree crowns of roadside trees, the vegetation canopy above the road, and the visible tree crowns that can form occlusion in the street view image. The bicycle lane surface category is used to label the bicycle lane surface or non-motorized vehicle passage surface that cyclists actually use. The sky opening category is used to label the sky portion in the street view image that is not obscured by tree crowns, buildings, or other obstructions. The shadow category is used to label the shadow areas formed by tree crowns, buildings, or other obstructions on the road surface, cycling space, or adjacent space. The background category is used to label other pixels that do not participate in the subsequent calculation of lateral misalignment, center trajectory sky opening, and reverse thermal toughness risk value.

[0134] Street view image samples are used as input to the image semantic segmentation model, and the pixel label maps corresponding to the street view image samples are used as supervision information to train the image semantic segmentation model. The image semantic segmentation model can be a semantic segmentation network with an encoder and decoder structure, a convolutional neural network semantic segmentation network, or a semantic segmentation network based on a visual Transformer. The input of the image semantic segmentation model is the three-channel image data corresponding to the street view image samples. The three-channel image data undergoes size unification and pixel value normalization processing before being input into the model. Size unification refers to adjusting street view image samples of different sizes to the same input size, or cropping the street view image samples into image blocks that meet the model input requirements. Pixel value normalization processing refers to converting the pixel values ​​of each channel in the street view image samples to a uniform numerical range, so that images from different acquisition devices or under different lighting conditions can be input into the model in a consistent data form.

[0135] The output of the image semantic segmentation model is a pixel category probability map corresponding to the size of the input street view image sample. This pixel category probability map includes tree canopy category probability, bike path surface category probability, sky opening category probability, shadow category probability, and background category probability. For any pixel in the street view image sample, the image semantic segmentation model outputs the probability value that the pixel belongs to the tree canopy category, bike path surface category, sky opening category, shadow category, and background category, respectively. These category probabilities are used to determine the final category of the pixel and further generate the tree canopy region, bike path region, sky region, and shadow region required for subsequent calculations.

[0136] During training, the pixel-level classification difference between the pixel category probability map and the pixel label map is used as the training loss to update the model parameters of the image semantic segmentation model, so that the predicted category corresponding to the same pixel gradually approaches the labeled category in the pixel label map. The training loss can be pixel-level cross-entropy loss or a combination of pixel-level cross-entropy loss and region overlap loss. Pixel-level cross-entropy loss is used to constrain the category prediction result of each pixel to be close to the corresponding labeled category, and region overlap loss is used to constrain the spatial distribution of the same category region output by the model to be consistent with the corresponding labeled region in the pixel label map. Through the above training, the image semantic segmentation model can learn the visual differences between tree canopy, bike path surface, sky opening, shadow and background in street scene images, and can output pixel-level classification results for street scene images of target cycling sections.

[0137] After training, street view images of the target cycling route will be used. The image semantic segmentation model is input, and the output of the image semantic segmentation model is the street view image. Consistent pixel category probability maps; for street view images For any pixel in the image, compare the probabilities of the tree canopy category, bike path surface category, sky opening category, shadow category, and background category. Determine the category with the highest probability as the pixel's category. If the pixel is classified as a tree canopy, add it to the tree canopy region. If the pixel is classified as a bike lane surface, then add it to the bike lane area. If the pixel is classified as a sky opening, then add it to the sky region. If the pixel is classified as a shadow, then add it to the shadow area. If the pixel is classified as background, then the pixel will not be added to the canopy region. Bicycle lane area Sky area and shaded areas Canopy regions are generated based on the probability maps of the same pixel category. Bicycle lane area Sky area and shaded areas The initial binary mask;

[0138] When the image semantic segmentation model outputs multiple class probabilities for the same pixel, and the multiple class probabilities have the same maximum value, the final class of the pixel is determined in the order of bike path surface class, tree canopy class, sky opening class, shadow class, and background class. This process ensures that each pixel has a unique class, thus providing a clear pixel basis for subsequent region intersection, pixel count, and lateral projection calculations. The bike path surface class is given higher priority in the class determination order to ensure that the bike path area is prioritized. The integrity of the data ensures a stable regional basis for subsequent centerline extraction, generation of the observation area above the center trajectory, and calculation of the surface shadow coverage of the bicycle lane.

[0139] The generated initial binary mask can be processed by removing small connected components and filling holes to obtain a binary mask for subsequent calculations. Specifically, for the canopy region... The initial binary mask is used to remove isolated pixel regions that are discontinuous with the main canopy region and have a significantly smaller area. Holes inside the canopy region caused by local missed segmentation are filled to maintain the continuity of the canopy region in the image space. For the bicycle path region... The initial binary mask retains the main connected regions corresponding to the cycling surface, removes isolated regions unrelated to the bike path surface, and fills local holes within the bike path area to ensure stable extraction of the bike path centerline later; for the sky area... The initial binary mask is used to remove isolated mis-segmented regions caused by reflections, signs, or localized highlights; for shadow areas... The initial binary mask removes small mis-segmented regions caused by local noise and preserves areas that can be segmented with the bike lane area. The intersecting shaded areas. The binary masks resulting from the above post-processing are then used as the canopy regions. Bicycle lane area Sky area and shaded areas This serves as the input for subsequent calculations of lateral misalignment, central trajectory sky opening, and reverse thermal toughness risk value.

[0140] To ensure the reliability of subsequent calculations, the canopy area Bicycle lane area Sky area and shaded areas All images are saved as pixel sets and are compared with street view images. Maintaining the same image coordinate system allows for pixel-by-pixel correspondence between regions when subsequently calculating lateral projection, region intersection, pixel count, and the observation area above the center trajectory.

[0141] The image semantic segmentation model output in this embodiment is not directly used as the conclusion of thermal toughness assessment, but rather as the basis for subsequent calculations of the lateral misalignment between the canopy area and the bike path area, the sky opening of the center trajectory, and the reverse thermal toughness risk value. Even if the canopy area accounts for a high proportion in the street view image, it is still necessary to perform joint calculations of the lateral misalignment, the sky opening of the center trajectory, and the degree to which the bike path is not covered by shadows in order to determine whether there is a reverse thermal toughness risk.

[0142] Example 3, based on Example 1, further illustrates the relationship between lateral misalignment, central trajectory sky opening, and reverse thermal toughness risk value;

[0143] For a street view image of the target cycling route If the image semantic segmentation result shows the canopy region The proportion is relatively high, but the canopy area is mainly located above the sidewalk on the left or right side of the image, while the bike path area... Located on the outer side of the motor vehicle lane or on the other side of the road edge, the canopy area and bicycle lane area The horizontally normalized projection has less overlap on the horizontal coordinate of the image, resulting in a smaller horizontal overlap and a smaller horizontal misalignment. The shade level is relatively high; even if the visible green coverage of the tree canopy is high at this time, it does not necessarily mean that the bicycle lane is effectively shaded.

[0144] Furthermore, if based on the bicycle lane area The observation area above the center trajectory corresponding to the extracted center line With the sky area If the intersection is large, then the sky opening of the central trajectory is large. The larger size indicates that there is still a large opening in the sky above the cyclist as they travel along the center line of the bike path, making it more likely that the cyclist will be exposed to direct solar radiation.

[0145] If the shaded area Bicycle lane area If the intersection of the two areas is small, the coverage of the shaded area within the bike path area is low, and the bike path is not fully shaded, indicating that the bike path surface does not form sufficient shade coverage. In this case, all four conditions must be met simultaneously: high canopy area ratio, high lateral misalignment, high sky opening at the center track, and high bike path not being shaded, to determine the reverse thermal toughness risk value. The increase in green visibility can accurately indicate that there is a risk of sun exposure on the cycling route at the sampling location despite the appearance of high green visibility.

[0146] Conversely, if the canopy area Bicycle lane area They are highly overlapping in the lateral projection, and their lateral misalignment is... Smaller, or the observation area above the center trajectory The mid-sky region occupies a relatively small area, and the sky opening at the center of the trajectory is small. If the area is small, or if the shadow coverage within the bike lane area is sufficient, then even if a canopy area exists in the street view image, the reverse thermal toughness risk value will be lower. It will not be significantly magnified; therefore, this embodiment can avoid simply judging all street scene images with more tree canopies as low risk, and can also avoid misjudging roads where tree canopies do not act on bicycle lanes as high thermal toughness roads.

[0147] Example 4, based on Example 1, further illustrates the method for generating the thermal toughness assessment results of the target cycling section;

[0148] The target cycling route includes multiple consecutive sampling locations, each corresponding to a street view image. For each street view image, the reverse thermal toughness risk value is calculated according to steps S1 to S5 to obtain a risk sequence. Since cyclists move along a continuous path in the road, although a high risk at a single sampling location can reflect local sun exposure problems, the impact on the thermal toughness of cycling traffic is greater when multiple consecutive sampling locations have high risks.

[0149] This embodiment not only calculates the cumulative reverse thermal toughness risk value of all valid sampling locations, but also introduces a continuous risk term between adjacent valid sampling locations. When both adjacent valid sampling locations have high reverse thermal toughness risk values, the continuous risk term increases accordingly, indicating that the cyclist is continuously exposed to the risk of sun exposure due to the misalignment between the visible green of the tree canopy and the shade line of the bike path between these two adjacent locations; when at least one of the two adjacent valid sampling locations has a low reverse thermal toughness risk, the continuous risk term decreases, indicating that the high risk does not extend continuously along the cycling direction.

[0150] By using the continuous exposure risk value Q, this embodiment can unify the single-point exposure risk and the continuous exposure risk along the road segment into the same evaluation value. This allows the thermal resilience assessment results to reflect both the local risks at certain streetscape sampling locations and the degree to which the risks persist along the cycling direction in the entire target cycling segment. The generated thermal resilience assessment results are more suitable for identifying vulnerable road segments in cycling traffic under extreme high temperatures, and are particularly suitable for identifying road segments where the streetscape greenery appears sufficient but the center track of the bicycle lane is continuously exposed to the sun.

[0151] like Figure 2 As shown, one embodiment of the present invention also provides an urban cycling traffic thermal resilience assessment system based on street view perception and machine learning;

[0152] In this embodiment, the functions of each module / unit of the system are as follows:

[0153] The image segmentation module is used to acquire a sequence of street scene images for thermal resilience assessment of the target cycling route. A trained image semantic segmentation model is used to segment each street scene image in the sequence, extracting the canopy region, bike lane region, sky region, and shadow region.

[0154] The trajectory generation module is used to extract the center line of the bicycle lane area and generate an observation area above the center trajectory in the street view image based on the center line.

[0155] The misalignment detection module is used to calculate the projections of the tree canopy area and the bicycle lane area in the horizontal direction of the street view image, and to calculate the lateral misalignment degree between the tree canopy area and the bicycle lane area based on the overlap relationship between the projections of the tree canopy area and the bicycle lane area.

[0156] The aperture calculation module is used to calculate the sky aperture of the central trajectory based on the intersection of the observation area above the central trajectory and the sky area.

[0157] The risk assessment module is used to calculate the reverse thermal toughness risk value of the street view image based on the proportion of the canopy area in the street view image, the coverage of the shadow area in the bicycle lane area, the lateral misalignment, and the sky opening of the center trajectory.

[0158] The resilience assessment module is used to calculate the continuous exposure risk value based on the reverse thermal resilience risk value of each street view image and the arrangement order of each street view image in the target cycling section, and to generate the thermal resilience assessment result of the target cycling section based on the continuous exposure risk value.

[0159] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning, characterized in that, Includes the following steps: A street view image sequence is acquired for thermal toughness assessment of the target cycling route. A trained image semantic segmentation model is used to segment each street view image in the street view image sequence to extract the canopy region, bike lane region, sky region, and shadow region. Extract the centerline of the bicycle lane area, and generate an observation area above the center trajectory in the street view image based on the centerline; Calculate the projections of the tree canopy region and the bicycle lane region onto the horizontal direction of the street view image, respectively. Based on the overlap relationship between the projections of the tree canopy region and the bicycle lane region, calculate the lateral misalignment between the tree canopy region and the bicycle lane region. The sky opening of the central trajectory is calculated based on the intersection of the observation area above the central trajectory and the sky area. The reverse thermal toughness risk value of the street view image is calculated based on the proportion of the canopy area in the street view image, the coverage of the shadow area in the bicycle lane area, the lateral misalignment, and the sky opening of the center trajectory. The specific steps for calculating the reverse thermal toughness risk value of the street view image are as follows: The ratio of the number of pixels in the tree canopy region to the total number of effective pixels in the street view image is calculated to obtain the proportion of the tree canopy region. The number of overlapping pixels between the bike lane area and the shadow area is obtained, and the ratio of the number of overlapping pixels to the total number of pixels in the bike lane area is calculated to obtain the coverage of the shadow area in the bike lane area. Subtract the coverage of the shaded area within the bike lane area from the value 1 to obtain the degree to which the bike lane is not covered by shadow; Multiply the canopy area ratio, the lateral misalignment, the sky opening of the central trajectory, and the degree to which the bicycle path is not covered by shadow to obtain the reverse thermal toughness risk value. Based on the reverse thermal toughness risk value of each street view image and the arrangement order of each street view image in the target cycling section, a continuous exposure risk value is calculated, and a thermal toughness assessment result of the target cycling section is generated based on the continuous exposure risk value.

2. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, Extract the canopy area, bike path area, sky area, and shade area, including: A trained image semantic segmentation model is used to perform pixel-level segmentation on each street scene image to obtain pixel sets belonging to tree canopy, bike path surface, sky opening and shadow respectively; The set of pixels belonging to the tree canopy is defined as the tree canopy region; The set of pixels belonging to the surface of the bicycle lane is defined as the bicycle lane region; The set of pixels belonging to the sky opening is defined as the sky region; The set of pixels that belong to the shadow is defined as the shadow region.

3. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, Extract the centerline of the bicycle lane area, and generate an observation area above the center trajectory in the street view image based on the centerline, including: The bicycle lane area is skeletonized to obtain the centerline representing the bicycle lane traffic position. For any pixel on the center line, the pixels located at the horizontal position of the pixel along the vertical direction of the street view image to the upper boundary of the street view image are included in the candidate region. All candidate regions corresponding to the center line are merged to obtain the observation area above the center trajectory.

4. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, Calculate the projections of the tree canopy region and the bike path region onto the horizontal direction of the street view image, respectively. Based on the overlap relationship between the projections of the tree canopy region and the bike path region, calculate the lateral misalignment between the tree canopy region and the bike path region, including: Calculate the normalized projections of the tree canopy region and the bicycle lane region along the lateral direction of the street view image, respectively; For each lateral coordinate of the street view image, determine the minimum value between the normalized projection of the canopy region and the normalized projection of the bicycle lane region; The minimum values ​​corresponding to all horizontal coordinates are summed to obtain the horizontal overlap. Subtract the lateral overlap from the value 1 to obtain the lateral misalignment.

5. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, The sky aperture of the central trajectory is calculated based on the intersection of the observation area above the central trajectory and the sky area, including: Obtain the total number of pixels within the observation area above the central trajectory; Obtain the number of pixels in the overlapping area between the observation region above the central trajectory and the sky region; Calculate the ratio of the number of pixels in the overlapping intersection region to the total number of pixels in the observation area above the central trajectory, and use the ratio as the sky aperture of the central trajectory.

6. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, Based on the reverse thermal toughness risk value of each of the street view images and the arrangement order of the street view images in the target cycling segment, the continuous exposure risk value is calculated, including: According to the order in which the sampling positions of each street view image are arranged in the target cycling segment, the reverse thermal toughness risk values ​​of each street view image are combined into a risk sequence. Calculate the cumulative sum of all reverse thermal toughness risk values ​​in the risk sequence; Calculate the square root of the product of two adjacent reverse thermal toughness risk values ​​in the risk sequence, and sum the square roots of all adjacent two reverse thermal toughness risk values ​​to obtain the sum of adjacent continuous risks. The cumulative sum is added to the sum of adjacent consecutive risks, and then divided by the normalization coefficient to obtain the continuous sun exposure risk value. The normalization coefficient is twice the total number of images in the street view image sequence minus one.

7. The method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning according to claim 1, characterized in that, Based on the continuous exposure risk value, a thermal resilience assessment result for the target cycling section is generated, including: The continuous exposure risk value is used as the reverse evaluation value of the thermal toughness of the target cycling section; The risk value of continuous sun exposure is negatively correlated with the thermal toughness of cycling traffic on the target cycling route.

8. A system for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning, used to execute the method for assessing the thermal resilience of urban cycling traffic based on street view perception and machine learning as described in any one of claims 1-7, characterized in that, include: The image segmentation module is used to acquire a sequence of street scene images for thermal resilience assessment of the target cycling route. A trained image semantic segmentation model is used to segment each street scene image in the sequence, extracting the canopy region, bike lane region, sky region, and shadow region. The trajectory generation module is used to extract the center line of the bicycle lane area and generate an observation area above the center trajectory in the street view image based on the center line. The misalignment detection module is used to calculate the projections of the tree canopy area and the bicycle lane area in the horizontal direction of the street view image, and to calculate the lateral misalignment degree between the tree canopy area and the bicycle lane area based on the overlap relationship between the projections of the tree canopy area and the bicycle lane area. The aperture calculation module is used to calculate the sky aperture of the central trajectory based on the intersection of the observation area above the central trajectory and the sky area. The risk assessment module is used to calculate the reverse thermal toughness risk value of the street view image based on the proportion of the canopy area in the street view image, the coverage of the shadow area in the bicycle lane area, the lateral misalignment, and the sky opening of the center trajectory. The resilience assessment module is used to calculate the continuous exposure risk value based on the reverse thermal resilience risk value of each street view image and the arrangement order of each street view image in the target cycling section, and to generate the thermal resilience assessment result of the target cycling section based on the continuous exposure risk value.

Images