1  Street Views

Why street views?

Street view images represent real-world scenery from a pedestrian perspective. By analyzing the elements in street view images, the urban built-environment can be better modelled. ¹

How did we collect the images?

Given the streets class by OpenDataPhilly, streets' class lower than minor arterial are not considered to exclude local pathes did we barely have street view images. Then, a set of points is selected from the streets with a distance of 100m(328ft). And street view images are collected according to these points throuth Bing Map Api.

How did we quantify the elements in these images?

An Image Segmentation method SegNet is used to analyzing the elements and their proportion in the images. The elements in the images are devided into different classes, like trees, roads, cars or sidewalks.

Questions we are going to explore

1. How does the street space quantitatively vary within Philadelphia?
2. Are there clustering for the street spaces?
3. What is the relationship between races and their associated living streets? Does it achieve equity and where does the difference exist?
4. What is the relationship between crime occurrence and the quantity of living street elements?
5. For specific crime types, which environment is easier to be effected?

2  Distribution of Elements

We first count the pixel number of each segment (each color), and calculate the proportion of each segment by dividing the total pixel number of them.

Then, we combined the output elements into ten categories (Green, Wall, Lives, Building, Infrastructure, Road, Sidewalk, Sky, Transportation, and Public service), which will make the analysis more concise and intuitive.

We plot each category of street view point in the map of Philadelphia, using color to represents the proportion of each category. The purple represents a low proportion and the yellow represents a high proportion. Clusters can be observed from most categories, especially in lives, road and green.

3  Clustering

Based on the street elements analysis outcome, KMeans Clustering Method is introduces to find concise descriptions of these elements. After attempts, we set the clustering number as four.

Interestingly, even though we did not include the geometry feature when doing the clustering analysis, it still represents spatial clustering feature. This may because of the administration unit division and community segregation. Adjacent streets will receive similar urban administration and attract similar population.

Street View Point Clustering Aggregated Lable by Block Group Based on above quantitative diagram. We attach labels to these four categories, which may describe the feature of these clustering. The labels are as follow.

1: High-density

This clustering mainly distributes at the center city which is the most prosperous area in Philadelphia, and it has more buildings, less green, and more openness (less wall and more road) compared to other areas.

2: Lush

This clustering mainly distributes at the west and mid-north Philadelphia. The features of this clustering are low building density, high street greening, and wide road.

3: Spacious

This clustering distributes around large natural parks. And besides this, it has relatively less buildings. These two indicated that it is easier to see the sky in the street, which also validated by the analysis. And also this category has more walls, less open to the public compared to other areas.

4: Townhouse

As this name, townhouse, indicates, the mainly building type within this clustering is townhouse. Therefore, this clustering has more greening, which is mainly made up of private lawns and gardens of townhouses, and less road.

4  Street Elements & Race

From the plot, we can see that when the percentage of the white are at the mid-range, the street will have more buildings, lives, public service and transportation. That means a mixed community will bring vitality to the street, and we should encourage the confusion. However, we also see the white-majority communities enjoy a more greening and open street while the other communities have less green, more wall, roads and transportation. Since we only analyze the street elements, there are many other factors not being taken into account and we can not simply draw the conclusion.

5 Correlation of Street Elements & Crime

A scatter plot is used to present the correlation between each street element and crime incident. To further examine the relationship, we ran an OLS regression, where the crime count is the dependent variable and the street components are the independent variables. This regression's aim is not to accurately predict the crime count, but to see the coefficient and the significance of each independent variable to the crime count. (Due to some confidence interval cross over "0", so even these Building, Road, and Sidewald variables have negative coefficient in the regression model, they display a positive relationship in the scatter plot)

The figure below describes the coefficient of each element, along with their confidence interval. Element such as lives, sky, public service, and green are significant. Specifically, elements lives and public service have a positive relationship with crime numbers, while sky and green have a negative relationship with crime numbers, which means that streets with more green and less people is correlated with less crimes.