Urban Informality

Online evaluation for urban informality

We aim to identify urban informality in street space in Chinese cities. We are welcoming your input for our online comparison "game" via http://zhudeng.top/StreetViewRanking/main.php >>


Understanding Urban Informality in Street Built Environment

Combining Manual Evaluation with Machine Learning in Processing the Beijing Old City’s Street-view Images

How to understand urban informality in street built environment? In the process of rapid urbanization, a large number of informal workers enter megacities. Therefore, urban streets have become a vital space for those lower entry-level industries undertaken by these groups. On the one hand, informal workers agglomerate on streets, and legally occupy streets with unregulated roadside business activities, changing daily life of the streets consciously or unconsciously. On the other hand, residents who live along the streets continuously change street built environment based on their own daily needs. 


The existing studies about urban informality in street built environment mainly use qualitative research method, combining interview and participatory observation with environment-behavior study method, and on-going field studies of particular streets. However, such method has its limitation when identifying and analyzing spatial distribution of urban informality in large-scale space. The emergence of big data derived from street-view images has provided a vital source of data for street studies. Manual evaluation and Machine Learning based on data derived from street-view images as a survey tool has proved to be an effective way of cognizing built environment. However, through cognizing street-view images, most of these studies focus on physical spaces or environmental aspects of built environment, very few of them tend to establish the relationship between social conditions and built environment. Besides, manual evaluation and machine learning have their drawbacks to identifying and analyzing street-view images. Therefore, this paper aims for two main study objectives. The first is to explore new methodology that can combine manual evaluation with machine learning as a way to resolve the drawbacks to the two individual methods. The second is to cognize urban informality by means of street-view images, in which social aspects of street built environment in large-scale space are revealed. Originally such information was not able to be surveyed. It is expected to improve the clarity in describing aspects of spatial distribution of urban informality. The result is useful for categorical street renovation in the future.  


The empirical research extracted data set from Tencent’s street-view images of 841 streets in the area of 62.5 km2 in Beijing old city. We obtained 1886 street-view locations with 4 images of each direction, 7,544 street-view images in total. After filtering outsome pictures manually, we arrived at a data set of 6,142 Hutong images that are relevant to the research topic.


Summarized from previous literatures, street informality embodies three aspects: changes brought about by informal workers; reconstructions based on individual demands without government’s guidance; changes resulted from daily-living behaviors, including peddling, mendicity, graffiti, street vendors, and façade reconstruction led by community.


However, these three representations of street informality are not practical for evaluation. This is because researchers cannot determine whether the street has been changed by informal employers and behaviors or not, and the difference between guided reconstruction and unauthorized reconstruction cannot be identified as well. Therefore, after careful field observation of streets in historic conservation areas in Beijing old city, we further developed the concept of street informality into operational guidelines to establish our evaluation system. In our research, street informality is graded in terms of five factors: billboards inconsistent with traditional style or material, façade reconstruction inconsistent with traditional style, vendors selling outdoor or randomly along the street, unauthorized construction additional to the houses along the street, and household objects placed outdoor by residents. Each image is graded by the number of factors identified, with a maximum score of 5 points and a minimum of 0 point.


Manual evaluation was conducted by two researchers, who first graded the 400 images randomly selected from the field observation,and then compared the results and discussed the ones with distinctively different scores, to unify the grading criteria followed by another round of grading. 6142 images were graded by these two researchers, in which 3572 were randomly selected to become the training set to train the machine learning model, while the rest of 2570 were used to compare the respective performances of manual evaluation and machine learning.


Through comparing the evaluation results conducted by the researchers and the model, pros and cons are identified in both evaluation methods. Firstly, the manual evaluation is better at accurately identifying informal factors, while machine learning model makes evaluation mostly based on physical features, like colors, shapes, etc., making it less effective in understanding overall environment. Especially when evaluating urban social space, which tends to be an issue, machine learning models cannot identify intangible factors. Secondly, although empirical-based manual evaluation is relatively accurate in most of the cases, human subjectivity is still inevitable especially when the grading system is cross-referenced with multiple factors. Errors caused by cognitive differences between scorers can further increase the error of machine learning model. Thirdly, it’s hard to conduct manual grading on a large amount of images, while machine learning model has great advantage under such circumstance.


The results of manual evaluation of street-view images were correlated with 841 streets in Arcmap, reflecting the spatial distribution of urban informality in street built environment in Beijing old city. Comparing the street-view images in relevant areas, the paper finds that the streets characterized with urban informality tends to have following features: firstly,there is more vendors on the street-level;secondly, street-level businesses are mostly petty trades;thirdly,there are more extensional buildings along the streets. By contrast, there are three types of street built environments which bear no sign of urban informality: firstly, former historical districts that have become new urban blocks through regeneration process; secondly, areas once inhabited by the privileged class in the past and left with large amount of historic heritages; thirdly, run-down laneways with no activity of street vendors and recreational space for local residents.

甘欣悦等 2018 时代建筑_街道非正规性.pdf
Adobe Acrobat Document 12.4 MB