KR20210067861A - Apparatus and method for evaluating walking safety risk, and recording media recorded program realizing the same - Google Patents

Abstract

A walking safety risk evaluating device and a method thereof and a recording medium in which a program for realizing the same is recorded are provided. According to an embodiment of the present invention, the walking safety risk evaluating device comprises: a collection unit that collects information on major environmental factors of a city; an extracting unit for extracting a physical element by applying an RCNN algorithm to information on the environmental factors; and an analysis unit that analyzes whether there is an accident risk during walking by applying statistical modeling to the physical element.

Description

Pedestrian safety risk assessment apparatus and method, and a recording medium on which a program for implementing the same is recorded

The present invention relates to a pedestrian safety risk assessment apparatus and method, and to a recording medium in which a program for implementing the same is recorded.

There is a limit to understanding the potential traffic accident risk on the road only with traffic accident data. In general, pedestrian traffic collisions tend to occur at intersections or curved roads where visibility is obstructed by curves in the road, stationary objects, parked vehicles, darkness, or other visibility limitations. For example, a vehicle driver may not notice a pedestrian who may be obstructed and obscured by an obstacle.

Accordingly, there is a need for a method for evaluating the potential traffic accident risk of pedestrians.

US Patent Publication No. US 2019/0236373 (published on Aug. 1, 2019) Republic of Korea Patent Publication No. 10-2017-0138225 (published on December 15, 2017)

The present invention is to solve the above problems, and provides a pedestrian safety risk assessment apparatus and method capable of evaluating a potential traffic accident risk for preventing traffic accidents on roads in the living area, and a recording medium in which a program for implementing the same is recorded. .

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the following description.

A pedestrian safety risk assessment apparatus according to an embodiment of the present invention for achieving the above object includes: a collection unit for collecting information on major environmental elements of a city; an extracting unit for extracting a physical element by applying an RCNN algorithm to the information on the environmental element; and an analysis unit for analyzing whether there is a risk for an accident risk according to walking by applying statistical modeling to the physical element.

In addition, the collecting unit may collect an image including at least one of a pedestrian, a vehicle, a signboard, a street tree, and a road facility as information on the main environmental element.

Also, the extractor may extract the physical element using scene labeling including a scene segmentation model and an object detection model.

Also, the extractor may derive an area ratio of the sky, green area, road, and sidewalk based on the scene division model.

Also, the extractor may derive the number of pedestrians, the number of vehicles, and the presence of traffic lights by the object detection model.

In addition, the analysis unit may analyze whether the risk is based on an average value of the area of the sky, green area, road, and sidewalk.

In addition, it may further include an evaluation unit for evaluating a spatial point for the risk perception.

And, it may further include a providing unit that provides a space design guideline based on the spatial point.

A pedestrian safety risk assessment method according to an embodiment of the present invention for achieving the above object includes: collecting information on major environmental elements of a city; extracting a physical element by applying an RCNN algorithm to the information on the environmental element; applying statistical modeling to the physical element to analyze whether it is a risk for an accident risk according to walking; evaluating spatial points for each type of risk perception; and providing guidelines for each type based on the spatial point.

The collecting may include collecting an image including at least one of a pedestrian, a vehicle, a signboard, a street tree, and a road facility as information on the main environmental element.

Also, the extracting may include extracting the physical element using scene labeling including a scene segmentation model and an object detection model.

Also, the extracting may include deriving an area ratio of the sky, green area, road, and sidewalk by the scene division model.

In addition, the extracting may include deriving the number of pedestrians, the number of vehicles, and the presence of traffic lights by the object detection model.

In addition, the analyzing may include analyzing whether it is a risk based on an average value of the area of the sky, green area, road, and sidewalk.

A program for implementing the walking safety risk assessment methods according to various embodiments of the present invention for achieving the above-described task may be recorded in a computer-readable recording medium recorded thereon.

Other specific details of the invention are included in the detailed description and drawings.

According to the present invention, it is possible to preemptively prevent a pedestrian accident by evaluating a potential traffic accident risk.

In addition, it is possible to predict whether there is a danger on the road in the living area, evaluate the spatial point for each type of danger by comparing it with the actual accident risk, and provide guidelines for each type.

1 is a diagram showing the configuration of a walking safety risk assessment apparatus according to an embodiment of the present invention.
2 is a view showing a part of Ulsan city in Korea used in an actual study case.
3 is a diagram showing an example of applying a scene segmentation model and an object detection model to a street view image;
4 is a diagram illustrating an example of a criterion for detecting an optimal number of clusters.
5 is a diagram illustrating an example of a cluster center value for each type of intersection.
6 is a diagram showing an actual image and an image generated by Labelme software.
7 is a diagram showing an example of spatial distribution of four types of intersections.
8 is a diagram illustrating an example of an intersection in which a street view image has an average ratio of (a) sky, (b) green space, (c) road, and (d) sidewalk.
9 is a diagram illustrating an example of an intersection type 1;
10 is a diagram illustrating an example of intersection type 2;
11 is a diagram illustrating an example of intersection type 3;
12 is a diagram illustrating an example of an intersection type 4;
13 is a diagram illustrating a sequence of a method for evaluating a walking safety risk according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, and only these embodiments allow the disclosure of the present invention to be complete, and common knowledge in the technical field to which the present invention belongs It is provided to fully inform the possessor of the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout.

Although first, second, etc. are used to describe various elements, components, and/or sections, it should be understood that these elements, components, and/or sections are not limited by these terms. These terms are only used to distinguish one element, component, or sections from another. Accordingly, it goes without saying that the first element, the first element, or the first section mentioned below may be the second element, the second element, or the second section within the spirit of the present invention.

The terminology used herein is for the purpose of describing the embodiments and is not intended to limit the present invention. As used herein, the singular also includes the plural unless specifically stated otherwise in the phrase. As used herein, "comprises" and/or "made of" refers to a referenced component, step, operation and/or element of one or more other components, steps, operations and/or elements. The presence or addition is not excluded.

Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used in the meaning commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in a commonly used dictionary are not to be interpreted ideally or excessively unless clearly defined in particular.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings.

1 is a diagram showing the configuration of a walking safety risk assessment apparatus according to an embodiment of the present invention.

Referring to FIG. 1 , an apparatus 100 for evaluating risk for walking safety according to an embodiment of the present invention may include a collection unit 110 , an extraction unit 120 , and an analysis unit 130 . In addition, the walking safety risk evaluation apparatus 100 may further include an evaluation unit 140 and a providing unit 150 .

More specifically, the walking safety risk assessment apparatus 100 according to an embodiment of the present invention is a risk that can recognize the risk of an accident through the collection unit 110 , the extraction unit 120 , and the analysis unit 130 . A predictive map may be built, and a design guideline for road pedestrian safety may be provided through the evaluation unit 140 and the provision unit 150 . Specific functions of each component will be described later.

The collection unit 110 collects information on major environmental elements of the city. Specifically, information on major environmental factors is collected from images of real cities and converted into data.

Main environmental elements are parts that can be distinguished from each other in an image of a city, and may include buildings, roads, sky, sidewalks, cars, pedestrians, etc., and various environmental elements may be mixed. For example, in the environmental elements of the road, various environmental elements such as a parked vehicle, a signboard, a street tree, and a road facility in addition to a pedestrian and a driving vehicle may be mixed.

Accordingly, the collection unit 110 may collect an image including at least one of a pedestrian, a vehicle, a signboard, a street tree, and a road facility as information on the main environmental element. Road facilities include various facilities installed on roads in which vehicles operate, such as roadways (roads), sidewalks (sidewalks), traffic lights, guide signs, and median strips.

In order to collect information on the environmental factors of the actual road, the city image is captured using a 360-degree camera, and the information is collected in consideration of regional specificities and functions such as building height, elevation design, and auditory information. can be collected

In addition, the captured image of the city may be a street view published on an accessible network. Specifically, since the street view image is composed of images having various angles and viewpoints (viewpoints), it is easy to collect information on major environmental elements.

The city's photographed image is very similar to the perception of real pedestrians recognizing the street-level walking environment, so that the urban environment elements recognized at eye level can be collected.

In addition, the collection unit 110 may select an evaluation item of the physical characteristics of an actual road, select an appropriate image according to a criterion suitable for the purpose of analysis, and collect information.

In addition, the collection unit 110 may store the collected information in a database for each environment element. That is, the collection unit 110 may select an appropriate image according to a predetermined criterion, collect information, and build an image database for each element.

The extraction unit 120 extracts the physical element by applying the RCNN algorithm to the information on the environmental element.

For example, the extraction unit 120 applies the RCNN algorithm in consideration of various environmental factors such as parked vehicles, signboards, street trees, road facilities, etc. in addition to the driving vehicle, and spatial combination of each environmental factor with respect to the actual urban environmental factors. Physical elements can be extracted by analyzing the composition.

Here, the RCNN (Region Convolution Neural Network) algorithm divides and analyzes the existing CNN model into independent regions.

The CNN (Convolution Neural Network) algorithm is used for image analysis, but the input data of the CNN algorithm is the signal strength of pixel data, and since the data classification standard is a black box model, it is used for the purpose of presenting specific guidelines for pedestrian safety. There is a problem that the limit of usability is clear.

Accordingly, an algorithm capable of object detection, which was a limitation of the CNN model, is the RCNN (Region Convolution Neural Network) algorithm.

By applying the RCNN algorithm by the extraction unit 120 to extract a physical element, it can be used as data of the traffic accident risk recognition map.

More specifically, the extractor 120 may extract a physical element using scene labeling including a scene segmentation model and an object detection model.

Scene labeling methods allow the identification and extraction of individual components of the entire scene, as well as complex relationships between entities commonly found in street scenes, such as pedestrians, cars, roads, or sidewalks.

For example, if scene labeling is applied to street images, it is possible to analyze the environmental characteristics that are measured and constructed at eye level. Eye level measurement can more accurately reflect a person's actual perception of the attributes of a particular environment.

Combining captured and collected street-level images with scene labeling methods can yield data about what people typically experience and perceive at eye level on the ground.

The ability of scene labeling to analyze the urban environment at eye level makes it possible to measure the visibility of pedestrians and drivers, one of the important elements of city traffic safety.

In general, pedestrian traffic accidents tend to occur at intersections or curves where visibility is obstructed by the curvature of the road, stationary objects, parked vehicles, darkness, or other visibility limitations, obstructing the driver's view of the vehicle and causing the driver to avoid obstacles. Pedestrians that are obscured by the can be obscured. Therefore, improving pedestrian and driver visibility at intersections is essential to raising safety awareness and reducing the risk of collisions.

Accordingly, the extraction unit 120 may derive a building, sky, green area, road, sidewalk, and the like by the scene division model. More specifically, it is possible to derive area ratios of the sky, green areas, roads, and sidewalks. The area ratio of each environmental element can be calculated by counting the number of pixels in the divided mask.

Also, the extractor 120 may derive the number of pedestrians, the number of vehicles, and the presence of traffic lights by the object detection model.

The analysis unit 130 analyzes whether it is a risk for an accident risk according to walking by applying statistical modeling to a physical element.

Risk perception is the perception of risk, and the pedestrian can make decisions to avoid undesirable consequences based on risk perception.

Studies on pedestrian collision risk using only recorded collision data pose difficulties in evaluating road safety due to the relatively lack of reliable data on pedestrian collisions. In this case, information on whether there is a risk can be used as an indicator of the potential risk of collision at a location where an actual collision does not occur. This is because an individual's awareness of the risk of collision provides useful additional information to proactively identify potential hazard locations or hazards.

The analysis unit 130 may construct a negative binomial or poisome statistical model by using the physical elements extracted through the RCNN algorithm. This makes it possible to construct a risk perception prediction map for the road using the simple method and the kernel method of the parameter statistical model. And, through spatial comparison with actual accident frequency, it becomes the basis for defining two types of spatial inconsistency points, which will be described later.

In addition, the analysis unit 130 may classify the extracted physical elements by type and analyze whether it is a risk in consideration of the average value of the area for each type. For example, the analysis unit 130 may analyze whether it is a risk based on an average value of the area of the sky, green area, road, and sidewalk.

In addition, the analysis unit 130 may analyze the effect of risk perception due to the density of vehicles and pedestrians. In other words, in order to evaluate spatial points by type for risk perception, the impact of risk perception can be identified by considering vehicle and pedestrian densities.

The evaluation unit 140 may evaluate a spatial point for risk perception. More specifically, the evaluation unit 140 may define two spatial discrepancy points through spatial comparison with the actual accident frequency.

For example, in a point with a low perceived risk and a high frequency of actual accidents, low attention to traffic safety is the main cause of accidents, and in a situation where the inconsistency problem persists, the frequency of accidents is likely to increase at the same point, It can be a point that requires the installation of traffic safety facilities for attention to safety issues.

In addition, if the point where the perceived risk is high and the actual accident frequency is low, the risk of future accidents may be underestimated if the point where the behavior of pedestrians is changed due to the perception of the risk is determined only from the data of the past accident frequency.

That is, the perceived risk of accidents and the actual frequency of accidents may be inconsistent, and the evaluation unit 140 separates a point where the perceived risk is low and the actual accident frequency is high, and a point where the perceived risk is high and the actual accident frequency is low in two spatial areas. It can be defined as the point of inconsistency.

Therefore, based on the analysis of whether the risk for the risk of an accident caused by walking by applying statistical modeling to the physical element is analyzed, the evaluation unit 140 may evaluate the spatial point for the recognition of the risk by comparing it with the actual risk of an accident. .

The providing unit 150 provides spatial design guidelines based on spatial points. More specifically, the providing unit 150 may perform a three-dimensional simulation of the space design guideline, perform virtual reality with the three-dimensional simulation as a background, and provide it to the user.

Accordingly, it is possible to provide appropriate and proactive information that can reduce the risk of a potential collision.

In the following, let's look at an actual research case.

2 is a view showing a part of Ulsan city in Korea used in an actual study case.

Referring to FIG. 2 , Ulsan, one of the largest industrial complexes in Korea, was selected, and eight elementary school districts were selected in the same district. Between 2010 and 2014, there were 792 pedestrian collisions reported by police. Land use within the chosen site is primarily residential, but includes commercial and industrial areas. The number of enrolled students during the 2015 survey period was 4,130.

A risk perception survey was conducted in July 2015, and the participants in the survey were students aged 10-12 who were enrolled in 8-12 elementary schools. The number of respondents was 1,263, and the questionnaire collected data on crash accidents along demographic characteristics, way to and from school, and the route children walked. Among 1,263 students, 1,042 students completed all answers to the questionnaire, and 799 students who walked to and from school were included as final participants, excluding 243 respondents who used cars for commuting.

Participants in the survey perceived that 2,621 locations in the target area were at high risk of collision. Since we focused on the intersection, the reported location should spatially coincide with that intersection. A road junction was defined as having a 15 m (50 ft) buffer area at the junction of two road centerlines, so the number of hazard points located at 546 junctions was used as the collision risk score for each junction.

Risk exposure, one of the most important factors to consider when determining the number of pedestrian collisions, must be accurately considered when analyzing risk factors for pedestrian safety. It is likely to evaluate a specific intersection to school.

The number of students crossing at each intersection was derived from the survey data. Participants were asked to indicate regular walking routes to the school on a local map. Based on this information, the number of individual intersections at each intersection chosen for the route was counted and used as one of the exposure variables. To calculate the population and street densities, the population and street centerlines of GIS were used from the Ulsan National Statistical Information Service (2010) and Statistics Geographic Information Service (SGIS) in Korea.

3 is a diagram illustrating an example of applying a scene segmentation model and an object detection model to a street view image.

Referring to FIG. 3 , physical elements were extracted using a scene segmentation model and an object detection model from a street view image (map.naver.com) of each intersection including pedestrians, vehicles, signboards, street trees, road facilities, and the like.

The physical elements extracted from the scene segmentation model include the ratio of areas of sky, greenery, roads and sidewalks. The area ratio of each physical element in the distance view image counts the number of pixels in the segmentation mask. By object detection model, the number of pedestrians, vehicles, and traffic lights are extracted as physical elements.

At this time, coordinates of 546 intersections examined were calculated using ArcGIS 10.4 to download the target street view image. Based on the coordinates of each intersection, we collect 2,184 street view images in 4 directions at 546 intersections. The collected street view image includes at least one of pedestrians, vehicles, signboards, street trees, and road facilities with the intersection as the center.

To reduce pedestrian vehicle collisions, three approaches can be categorized: improving visibility, managing vehicle speed, and separating pedestrians from vehicles. Accordingly, three types of environment properties may be defined. This includes visibility of pedestrian/drivers, traffic flow, and separation of pedestrians from vehicles.

Visibility of pedestrians/drivers can be measured using a proportional area of sky and green. As shown in FIG. 3 , as the sky increases, the FOV of pedestrians and drivers increases, and as the number of green colors increases, the FOV tends to decrease. The sky and a little bit of green often limit the distance between drivers and pedestrians.

To quantify the traffic flow at the intersection, we can use the proportional area of the road extracted from the scene segmentation model and the number of pedestrians and vehicles identified from the object detection model.

Wider lane widths and high traffic are associated with greater risk of collisions. Conversely, in the case of a street section with sidewalks, installation of traffic lights, separation of pedestrians from vehicles, etc., the risk of collision between vehicles and pedestrians can be reduced.

The separation of pedestrians and vehicles can be expressed using the proportional area of the sidewalk and the presence of traffic lights extracted from the scene segmentation model.

4 is a diagram illustrating an example of a criterion for detecting an optimal number of clusters.

Intersections with a high degree of openness or a high percentage of sky may have low building density. In general, the wider the road, the higher the average building height. Many urban spaces can be classified into certain types of urban forms according to their similarities.

In particular, the street view may be classified into several groups. We use empirical cluster analysis to understand the distinct changes in the building environment between groups. It is a method of combining observations into groups based on similarity within a predetermined set of characteristics.

Five physical elements extracted from a scene segmentation model using K-means cluster analysis: building, sky, green space, road, and sidewalk ), we can classify intersections into different types based on their similarity.

Referring to FIG. 4 , 533 intersections were classified based on the similarity of values of five physical elements extracted from the scene segmentation model using K-means cluster analysis, and the four-cluster solution was the most It was a good cluster solution. STATA software (version 13.0) was used for cluster analysis.

5 is a diagram illustrating an example of a cluster center value for each type of intersection.

Referring to FIG. 5 , there are shown four types of intersections according to the four clusters, which are a wide road with sidewalks (type 1), a wide road with trees (type 2), and a wide road with clear visibility (type 3), narrow alleyways with dense buildings (Type 4).

Specific cluster center values are shown in Table 1 below.

Here, the centroid represents the intersection features representing the characteristics of each intersection type.

To investigate environmental attributes and their effects on perceived safety, two negative binomial (NB) regression analyzes were performed. Compared with the Poisson regression model, the standard negative binomial model is the most widely applied in traffic safety research because it can account for the overdispersion of collision changes and heterogeneity between locations.

The standard negative binomial model may have a probabilistic structure as shown in Equation 1 below.

는 교차료 i에서 포아송 평균

에 대한 조건이 포아송 분포되고 모든 교차로에 대해 독립적일 때 교차점 i에서 인지된 충돌 위험(perceived crash risk)이다.From here,

is the Poisson mean at intersection i

It is the perceived crash risk at intersection i when the condition for is Poisson distributed and independent for all intersections.

The Poisson mean can be expressed as Equation 2 below.

는 감마 분포(Gamma distribution)에 따른 모형의 곱셈 랜덤 효과이며,

는 공변량의 함수이고,

는 환경 속성의 벡터이며, β는 알려지지 않은 계수의 벡터이다.From here,

is the multiplicative random effect of the model according to the gamma distribution,

is a function of the covariate,

is a vector of environmental properties, and β is a vector of unknown coefficients.

We created two models for defining environment properties. To investigate the association between the independently constructed environmental features and the perceived risk of collision, and to focus on the association between the type of intersection and the perceived risk of collision. The NB regression model was estimated using STAB software (version 13.0).

To test the validity of the applied algorithm, a random sample of distance view images (434/2132) was used as a validation set to estimate model performance. For the quantitative evaluation of the algorithm, the results obtained from the scene segmentation algorithm were compared with the ground truth, and then the result score was calculated.

The precision score can be calculated using Equation 2 below.

6 is a diagram showing an actual image and an image generated by Labelme software.

The ground truth was annotated manually using Labelme (github.com/wkentaro/labelme), a polygon image annotation tool written in Python.

Referring to FIG. 6 , there are an image (right) generated by the Labelme software and an actual image (left) that is empirical data. To test the validity of the object detection model, the number of manually calculated environment properties and the number of environment properties detected by the object detection model were compared, and then weighted kappa coefficients were calculated for the quantitative evaluation of the algorithm.

The estimated performances of the scene segmentation model and the object detection model are shown in Table 2 below.

Referring to Table 2, we show that the scene segmentation model performed well in classifying variables (sky and road) that occupy a large area in street view images, but tends to have relatively low accuracy when classifying greenery and sidewalks. The average ratio of green and sidewalk in street view images is about 5% and 3%, respectively.

The object detection model showed moderate performance when detecting pedestrians (kappa 0.521), vehicles (kappa 0.676), and traffic lights (kappa 0.554).

7 is a diagram showing an example of spatial distribution of four types of intersections. 8 is a view showing an example of an intersection in which the street view image has an average ratio of (a) sky, (b) green area, (c) road, and (d) sidewalk.

Table 3 below provides statistics on the variables used in the analysis.

=0.836). 각 교차로에서 하늘, 녹지, 도로 및 보도의 비례 영역의 평균 값은 28 % (SD = 11 %), 5 % (SD = 5 %), 14 % (SD = 5 %) 및 3 % (SD = 2 %)이다. As shown in Fig. 7, each type of intersection is evenly distributed in the field. The perceived risk of collision at an intersection was average 3.09 (SD=6.35 standard deviation). The average of students crossing one day at a crossing is 4.75 (SD=9.08). The maximum number of students at the crossing is 66 and there are 215 unselected intersections on the way to school. There was a strong correlation between the risk of crashes and the number of intersections at intersections (

= 0.836). At each intersection, the average values of the proportional areas of sky, greenery, roads and sidewalks were 28% (SD = 11%), 5% (SD = 5%), 14% (SD = 5%), and 3% (SD = 2). %)to be.

In Fig. 8, the street view image shows the intersection point with the mean value of each variable in the field.

9 is a diagram illustrating an example of an intersection type 1; Also, FIG. 10 is a diagram showing an example of type 2 of the intersection. 11 is a diagram showing an example of intersection type 3; 12 is a diagram showing an example of type 4 intersection.

Table 4 below presents the parameter estimates and p-values of the two NB models for the perceived risk of collision.

and variance parameter estimates α were significantly higher than zero in both NB model 1 (95% CI (0.88, 1.37)) and NB model 2 (95% CI (0.89, 1.38)). In both models, all variables representing risk exposure were positively associated with perceived risk of collision. The higher the number of crossing students, the higher the population density, and the higher the street density, the higher the risk of cognitive crashes.

In NB model 1, the proportional area of the sky significantly reduced the perceived risk of collision (p = 0.023). Higher percentages of the sky indicate more visual openness than on the street, and areas more visible to drivers and pedestrians. For this reason, the negative relationship between the percentage of sky and the perceived risk of collision means that improved visibility at intersections may be one of the factors that reduces the perceived risk of collision.

Although the direct relationship between low visibility and actual risk of collision is unclear, improving pedestrian visibility at intersections may improve safety and alter pedestrian behavior.

On the other hand, the results of NB Model 1 showed that the proportional area of the road had a significant positive correlation with the perceived crash accident (p = 0.023). The width of lanes and roads is one of the important factors that increase the risk of collisions, and in fact, the wider the lane, the higher the vehicle speed, which increases the likelihood of collisions with pedestrians and serious injuries to pedestrians.

Pedestrian number was positively associated with perceived crash risk level (p = 0.071). This may be because more pedestrians at intersections impair the ability of school-aged children to cross. A variable in the number of pedestrians may indicate congestion at an intersection.

In Table 4 above, the differences between clusters in NB model 2 are shown.

Students tended to perceive lower risk at intersections with wide roads and clear visibility (Type 3) than at intersections with wide roads and sidewalks (Type 1; p = 0.030).

9 and 11, a clear distinction between the two types of intersections was the relative visual openness (eg, proportional area of the sky) and the presence of sidewalks. It is analyzed that increasing the visibility of pedestrians and the presence of sidewalks can reduce the perceived risk of collision.

The risk perception at the reference intersection, Type 1, was not statistically different from the risk perception at the wide road and tree intersection on the main road. Roadside trees can enhance the landscape of the street view (see FIG. 10), but do not alter the perceived level of safety. In fact, roadside trees are both visual and physical barriers in traffic safety situations.

Through a comparison of type 1 (see FIG. 9) and type 4 (see FIG. 12) intersections, school-age children reported a lower risk of cognitive conflict at type 4 intersections than at type 1 intersections (p = 0.071). Sidewalks and visual openness can reduce the risk of crashes.

Reduced pedestrian visibility, wider roads, more students, higher population density, and higher street density were all significant, all associated with a higher risk of cognitive crashes at intersections.

By collecting street view images built at eye level, scene labeling techniques can be used to analyze their impact on perceived risk of collision. Ultimately, this approach can provide appropriate and proactive interventions that reduce the risk of possible collisions at intersections.

13 is a diagram illustrating a sequence of a method for evaluating a walking safety risk according to an embodiment of the present invention.

Referring to FIG. 13 , the pedestrian safety risk assessment method according to an embodiment of the present invention collects information on major environmental elements of a city (S10), and applies an RCNN algorithm to the information on the environmental elements to obtain physical elements. extract (S20), apply statistical modeling to the physical element to analyze whether it is a risk for an accident risk according to walking (S30), evaluate the spatial point by type for the risk perception (S40), and the spatial point Based on the guideline for each type is provided (S50).

When information on major environmental elements of the city is collected ( S10 ), an image including at least one of pedestrians, vehicles, signboards, street trees, and road facilities may be collected as information on the major environmental elements.

The image is an image of a city, and an image captured using a 360-degree camera or the like can be used. For example, it may be a street view published to an accessible network. As an example, an image DB of road intersections and segments may be constructed using naver street view.

In addition, an image database can be built by grouping roads in consideration of land use, building density, and road characteristics. For example, a wide road with sidewalks, a wide road with trees, a wide road with clear visibility, a narrow alley with dense buildings, etc.

For example, each image is classified according to whether it includes a building, sky, green space, road, sidewalk, etc. corresponding to major environmental elements that can be included in the image. and can be converted into a DB for each classified image group.

When the physical element is extracted by applying the RCNN algorithm to information on the environmental element ( S20 ), the physical element may be extracted using scene labeling including a scene segmentation model and an object detection model. Here, the RCNN (Region Convolution Neural Network) algorithm divides and analyzes the existing CNN model into independent regions.

In this case, the area ratio of the sky, green area, road, and sidewalk may be derived by the scene division model. The area ratio of each physical element may be derived by calculating the number of pixels in the division mask.

In addition, the number of pedestrians, the number of vehicles, and the presence of traffic lights can be derived by the object detection model. In this case, a region of interest (ROI) may be set in the image and an object to be detected may be extracted. For example, after specifying the initial search window area at the initial position of the object shown in the image, calculate the color probability distribution, find the central pixel of the object inside and outside the search window area, and then compare the two to converge. Returns the position and reduces the size of the search window. The size of the object can be estimated by estimating the pixels of the object by repeatedly performing the above steps, converting it to an actual scale using a correction table, and outputting it.

In addition, there is a method using a learning machine such as AdaBoost or SVM (Support Vector Machine) as well as RCNN when detecting an object, and a non-learning method using vector similarity of extracted features. Learning methods and non-learning methods can be appropriately selected and used according to the complexity of the object and background. For example, as a local feature of an image, a Haar-like feature using the difference of the sum of pixel (pixel) values between two or more adjacent blocks, or the sum of weight products using weights. feature) can be applied. In order to obtain the difference of the sum of pixel values between adjacent blocks when extracting Haral-like features, a mask considering a simple rectangular feature may be used.

When statistical modeling is applied to physical factors to analyze whether the risk for an accident risk according to walking is analyzed (S30), the risk can be analyzed based on the average values of the areas of the sky, greenery, roads and sidewalks.

At this time, a negative binomial or poiosis statistical model can be constructed using the extracted physical elements, and a predictive map of risk to the road can be constructed using the parameters and kernel method of the statistical model. have.

In addition, the effect of risk perception by vehicle and pedestrian density can be analyzed. In other words, in order to evaluate spatial points by type for risk perception, the impact of risk perception can be identified by considering vehicle and pedestrian densities.

Based on this risk awareness, a risk awareness map can be developed for a preemptive response to prevent road traffic accidents. And, it is possible to preemptively prevent pedestrian accidents by systematically comparing and analyzing this with the actual accident frequency.

When the spatial point for each type is evaluated for risk recognition (S40), the spatial point for each type may be evaluated based on the risk recognition due to the density of vehicles and pedestrians.

In this case, the spatial point for each type may be a spatial inconsistency point of two types through spatial comparison with the actual accident frequency.

Specifically, the perceived risk of accidents and the actual frequency of accidents may become inconsistent, and the points where the perceived risk is low and the actual accident frequency is high and the point where the perceived risk is high and the actual accident frequency is low are two types of spatial inconsistency points. can be

When the guideline for each type is provided based on the spatial point ( S50 ), the spatial design guideline for each type may be provided according to the spatial inconsistency point corresponding to the spatial point.

Accordingly, when using a VR-based walking simulator, the evaluator can directly walk around the target site in virtual reality and interact with the surrounding non-fixed landscape elements (vehicles, people, light, sound, etc.) to conduct a realistic walking safety risk assessment. have.

That is, it is possible to design a three-dimensional simulation through the spatial design guideline, and provide the evaluator to perform virtual reality with the three-dimensional simulation as a background.

Here, for road pedestrian safety, guidelines for not only space design but also operational aspects may be presented.

On the other hand, the walking safety risk assessment apparatus and method according to an embodiment of the present invention can be implemented as one module by at least one of software, firmware, and hardware, and the above-described embodiments of the present invention are programs that can be executed on a computer It may be implemented in a general-purpose computer that operates the program using a computer-readable recording medium. The computer-readable recording medium is implemented in the form of a magnetic medium such as a ROM, a floppy disk, a hard disk, an optical medium such as a CD or DVD, and a carrier wave such as transmission through the Internet. In addition, the computer-readable recording medium is distributed in a computer system connected to a network so that the computer-readable code can be stored and executed in a distributed manner.

Although embodiments of the present invention have been described above with reference to the accompanying drawings, those of ordinary skill in the art to which the present invention pertains can realize that the present invention can be implemented in other specific forms without changing the technical spirit or essential features. you will be able to understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

100: pedestrian safety risk assessment device
110: collection unit
120: extraction unit
130: analysis unit
140: evaluation unit
150: provider

Claims (15)

a collection unit that collects information on major environmental factors of the city;
an extracting unit for extracting a physical element by applying an RCNN algorithm to the information on the environmental element; and
and an analysis unit for analyzing whether there is a risk for an accident risk according to walking by applying statistical modeling to the physical element. The method of claim 1,
The collection unit,
Pedestrian safety risk assessment device that collects images including at least one of pedestrians, vehicles, signboards, street trees, and road facilities as information on the major environmental elements. The method of claim 1,
The extraction unit,
A pedestrian safety risk assessment apparatus for extracting the physical element using scene labeling including a scene segmentation model and an object detection model. 4. The method of claim 3,
The extraction unit,
A pedestrian safety risk assessment device for deriving the area ratio of the sky, green area, road and sidewalk by the scene segmentation model. 4. The method of claim 3,
The extraction unit,
A pedestrian safety risk assessment device for deriving the number of pedestrians, the number of vehicles, and the presence of traffic lights by the object detection model. The method of claim 1,
The analysis unit,
A pedestrian safety risk assessment device that analyzes the risk perception based on the average values of the areas of the sky, greenery, roads and sidewalks. The method of claim 1,
Further comprising an evaluation unit for evaluating a spatial point with respect to the risk perception, walking safety risk assessment device. The method of claim 1,
Further comprising a providing unit for providing a spatial design guideline based on the spatial point, pedestrian safety risk assessment device. collecting information on key environmental factors of the city;
extracting a physical element by applying an RCNN algorithm to the information on the environmental element;
applying statistical modeling to the physical element to analyze whether it is a risk for an accident risk according to walking;
evaluating spatial points for each type of risk perception; and
A walking safety risk assessment method comprising the step of providing a guideline for each type based on the spatial point. 10. The method of claim 9,
The collecting step is
Pedestrian safety risk assessment method comprising the step of collecting an image including at least one of a pedestrian, a vehicle, a signboard, a street tree, and a road facility as information on the main environmental element. 10. The method of claim 9,
The extraction step is
and extracting the physical element using scene labeling comprising a scene segmentation model and an object detection model. 12. The method of claim 11,
The extraction step is
Including the step of deriving the area ratio of the sky, green area, road and sidewalk by the scene segmentation model, pedestrian safety risk assessment method. 12. The method of claim 11,
The extraction step is
Including the step of deriving the number of pedestrians, the number of vehicles, and the presence of a traffic light by the object detection model, pedestrian safety risk assessment method. 10. The method of claim 9,
The analyzing step is
A pedestrian safety risk assessment method comprising the step of analyzing whether the risk is based on the average value of the area of the sky, green space, road and sidewalk. A computer-readable recording medium in which a program for implementing the method of any one of claims 9 to 14 is recorded.

Images