Decode Streets & Emotions
An environmental psychology research project exploring the relations between street view elements and human emotions applying brainwave signals and deep learning tools.
View full report here.
OVERVIEW
Rapid urbanization in China is transforming the lives of millions of people with more and more convenience. However, urban designers and planners failed to take the emotional feelings of the pedestrians into consideration due to the out-of-date methodologies when conducting preliminary research. We introduced a new urban environment evaluation method applying brainwave sensors and analyzed the data with the help of a variety of technical tools, generating distinctive insights for the realm of environmental psychology and urban studies.
Collaborated With
Chen Chaoran, Duarte Torre do Valle,
Olga Bialczak
Keywords
Environmental psychology, data analysis
Duration
Advisor
16 Weeks
Ercüment Gorgul, Bea Camacho
My Tasks
Methodology research, Experiment design, Data analysis
"It will be very interesting to see how this method is applied more widely in urban design...you know, it's cool to see how your brain will react to the world."
-Voluntary subject, Dec. 10, 2017
INITIAL RESEARCH
UNFRIENDLY URBAN SPACE
"Over the past 30 years, we eventually build a notorious urban environment with despicable inconvenience and dissatisfaction..."
-Prof. Dihua Li,
Landscape Apartment of Peking University
The discrepancy between urban and lanscape design and the walking experience of the pedestrians has escalated througout the process of urbanization in China. The problems of the urban environment emerge in both human and city scale. One can dive into the myriads of methods to evaluate a built environment, yet few are related to what pedestrians really feel.
SURVEY RESULTS
Factors that designers care the most
Ways of collecting public feedback
SECONDARY RESEARCH
Traditional Validation Methods
Surveys
Surveys
Photographs
Subjects are unable to give pertinent answers without being in a specific context.
Hard to represent the dynamic experience when walking in the environment.
An effective way to collect insights but biased answers are also expected.
State-of-art Validation Methods
Physiologial
Evaluation
Single Metric
Evaluation
Integrated Evaluation
Depending on bird-view observations, only providing hypothetical analysis for preliminary design.
Lengthy research processes, difficult for designers to learn and evaluate an existing built environment.
A new method to figure out how pedestrians are responding to the environment.
Pain Points Summary
There is a variety of evaluation methods for the urban environments, yet the traditional methods focus on the static images, not the dynamic experience, or not engaging the pedestrians in the process.
Key Take-aways
1. Research on the objective metric related to the subjects' feelings.
2. Identify the key factor causing positive/negative effects on the subjects.
3. Rule out the "improvised" feedbacks and other noise.
4. Attain feedbacks and statistics in an efficient and reliable manner.
How might we assess the urban environments according to people's feelings objectively?
EXPERIMENT DESIGN
SETTING-UP
According to the survey, the emotions of the pedestrian can be an objective metric to evaluate street qualities. Meanwhile, more than 80% of environmental information is related to visual perception. To quantify the visual inputs and the emotions, EEG sensors and video recorders were applied as input devices.
FIELD RESEARCH PROCESS
1. Target Site
North Xiangyang Road was selected as the research site. We divided it into 19 segments with 20 observation points evenly, which will be the benchmarks of retrieving data.
2. Recruit Subjects
23 pedestrians were randomly selected at the intersection and been informed of the experiment purpose on the date of Dec. 12, 2017.
3. Preparations
The voluntary subjects were asked to walk on the street wearing a head-mounted EEG signal device before relaxing for 30 seconds with eyes closed to naturalize brainwave data.
4. Walk & Record
They are also required to take a video of whatever might interest them while walking on the street silently. The video footage and the EEG data were then retrieved and aligned.
DATA PROCESSING
DATA FLOW
The photos were retrieved from the video according to the observation points and were processed into a color-categorized picture using a machine learning tool Bayesian Segnet, developed by Cambridge University to identify different elements, whose ratios were then calculated by Color Summarizer.
1. Bayesian Segnet: An open-source semantic segmentation tool : https://mi.eng.cam.ac.uk/projects/segnet/
2. Color Summarizer: An online tool for image color calculation: http://mkweb.bcgsc.ca/color-summarizer/?
3. EEG TGAM Module: A brainwave sensor ASIC module created by Neurosky: https://store.neurosky.com/products/eeg-tgam
BRAINWAVE SIGNALS & EMOTIONS
Meditation measures how calm and clear-minded you are at the moment. It indicates the level of mental "calmness" or "relaxation" with algorithm value from 0-100 outputting every second. The more your mind relaxes, the higher the algorithm output value.
Meditation Level
Attention measures how focused or single-minded you are at the moment. It indicates the intensity of mental "focus" with algorithm value from 0-100 outputting every second. The more you focus, the higher the algorithm output value.
Attention Level
IMAGE PROCESSING
DATA ANALYSIS
Preliminary Findings
The EEG signals and the ratios of elements were aligned in Tableau, a data visualization software to compute a heatmap of coefficients above, which describes the degree to which these two groups of data are correlated. In the bottom two rows, the coefficient between tree element and meditation signal is rather high which in accordance with the common belief that natural scenes make people more relaxed. On the attention signal, the road has a high positive effect and pavement has a more negative effect compared with other elements. The heatmap above demonstrates some preliminary insights based on the data. However, most correlations between the brainwave signals and the elements are subtle, indicating that a further study on the correlation between these data is required.
REGRESSION MODEL OPTIMIZATION
CONCLUSION & FUTURE WORK
COEFFICIENTS FROM LASSO REGRESSION MODEL
Conclusions
1. Although the correlation between street views and pedestrians’ emotions might vary due to the individual differences, specific street elements exert different subtle influences on people’s emotions.
2. Bike, road and pedestrian have a positive effect while pavement has a negative effect on subjects’ attention signal intensity. Building, vehicle and tree have no apparent correlation with the attention signal intensity.
3. The tree has a rather strong positive effect while pedestrian, pavement and road have a negative effect on subjects’ meditation signal intensity. Building, vehicle and bike have no apparent correlation with the meditation signal intensity.
Based on the regression model, a plausible explanation for the result is that bikes and passers-by provoke the attention or alertness of the subjects. Pavement reduces the attention intensity as it arouses the sense of safety of the pedestrians.
Future Work
For the future work, since we also take the dynamic visual factors on the street into consideration, we would like to add more elements into our variables such as sound and smell, which could make the result more comprehensive and adaptive.