From Trolley Stops to Walkable Blocks: Machine Learning Insights into San Diego’s Urban Fabric

Overview

This project analyzes walkability around trolley stations in downtown San Diego, leveraging machine learning techniques on street-level imagery. The goal is to enhance urban planning by identifying areas with varying levels of walkability using methods like semantic segmentation and spatial analysis. The full jupyter notebook for this project can be accessed here. Due to the size limitations of GitHub, you need to download the layers folder, which contains the necessary shapefiles to run the script, which can be accessed here.

Team Members

• Daniel Pryke
• Drew Suthipongchai
• Farrel Azhar

Abstract

The study integrates spatial analysis with visual classification techniques to assess and enhance urban walkability around trolley stops in downtown San Diego. Using Google Street View images, we perform network analysis to define catchment areas and evaluate walkability based on various factors.

Background and Methodology

Our methodology draws from key literature on urban walkability and includes several datasets:

1. Sidewalk Inventory Polyline Dataset: Provides representations of sidewalks, including their condition and construction year.

2. Google Street View Static API: Used to scrape street-level imagery for visual walkability analysis, processed with a semantic segmentation model.

3. ROAD_ALL Dataset: Comprehensive roadway information crucial for analyzing pedestrian routes and accessibility.

Identifying Study Area with Network Analysis

We conducted a walking distance network analysis for three trolley stations: Balboa Avenue Transit Center, Clairemont Drive, and Tecolote Road. Catchment areas were defined based on 5-minute, 10-minute, and 15-minute walking distances, focusing on the 10-minute interval for practical walking times.

Image Scraping with Google API

We scraped street-level imagery around trolley stations using the Google Static Street View API, downloading images at over 400 unique points. About 900 high-quality images were retained for analysis.

Image Download Function

def download_images(lat, lng, prefix):
    """Download street view images if available."""
    headings = [0, 90, 180, 270]  # North, East, South, West
    downloaded_files = []
    start_time = time.time()  # Record the start time
    for heading in headings:
        filename = os.path.join(input_directory, f"{prefix}_heading_{heading}_lat_{lat}_lon_{lng}.jpg")
        params = {
            "size": "1080x1080",  # Max resolution for the free tier
            "location": f"{lat},{lng}",
            "heading": heading,
            "key": API_KEY
        }
        response = requests.get("https://maps.googleapis.com/maps/api/streetview", params=params)
        if response.status_code == 200:
            with open(filename, 'wb') as f:
                f.write(response.content)
            downloaded_files.append(filename)
        else:
            print(f"Failed to download image at {lat}, {lng} with heading {heading}")
    end_time = time.time()  # Record the end time
    duration = end_time - start_time  # Calculate the duration
    print(f"Time taken to download images for point ({lat}, {lng}): {duration:.2f} seconds")
    return downloaded_files

DeepLab Model Visualization

This section demonstrates how to run the DeepLab model on a sample image and visualize the segmentation results. The following code snippet loads a sample image, performs segmentation using the model, and displays the results.

SAMPLE_IMAGE = 'mapilary.jpeg'

print('Running DeepLab on the sample image...')

def run_visualization(SAMPLE_IMAGE):
    """Inferences DeepLab model and visualizes result."""
    original_im = Image.open(SAMPLE_IMAGE)
    seg_map = MODEL.run(original_im)
    vis_segmentation(original_im, seg_map)

run_visualization(SAMPLE_IMAGE)

Sample Output Image

Adjusted Walkability DataFrame

The final DataFrame created from the semantic segmentation of street images includes various adjusted factors impacting walkability:

adjusted_street_greenery	adjusted_visual_enclosure	adjusted_dh_ratio	adjusted_obstacles	adjusted_visual_complexity	adjusted_sidewalk	geometry	adjusted_walkability_index
1.731673	0.239006	-0.801196	-0.602550	0.0	-0.338766	POINT (-117.21786 32.77504)	0.228166
2.071793	0.161678	-1.272547	-0.406234	0.0	0.660505	POINT (-117.21705 32.77436)	1.215196
1.756792	0.208461	-0.954633	-0.575025	0.0	-0.700005	POINT (-117.21648 32.77395)	-0.264410
-0.065953	0.353069	0.800694	-0.286263	0.0	-1.040573	POINT (-117.21420 32.76170)	-0.239025
2.015237	0.223784	-0.521862	-0.560223	0.0	-0.504752	POINT (-117.21332 32.76740)	0.652184
...	...	...	...	...	...	...	...

Data Analysis

Using QGIS, we conducted raster analysis to visualize and quantify walkability. We created cost-distance maps that reveal how walkability scores correlate with accessibility to trolley stations. These maps also highlight significant trends, such as the disparity in walkability between gridded neighborhoods and cul-de-sacs.

Key Findings

The adjusted walkability index indicates critical insights about urban fabric and infrastructure around trolley stations. For example, the Clairemont Drive Trolley Station demonstrates notably poor walkability, which reduces its effective catchment area. Conversely, neighborhoods designed with a gridded layout tend to exhibit higher walkability scores, emphasizing the role of urban design in facilitating pedestrian movement.

Visual Insights

Top and Bottom 5 Adjusted Walkability Index Locations

The following plot illustrates the geographic locations of the top 5 and bottom 5 adjusted walkability index points in downtown San Diego:

Discussion

The results underscore the necessity for targeted urban planning interventions. Our analysis provides a replicable framework for assessing urban walkability, making it a valuable resource for city planners and policymakers. Future research should focus on validating these findings through ground-truthing and expanding the geographic scope of the study to include additional trolley stations.

Conclusion and Future Work

This project not only sheds light on walkability around San Diego's trolley stations but also sets the stage for future explorations in urban planning. Potential next steps include:

• Validating model predictions with real-world observations.
• Utilizing higher-resolution imagery for enhanced analysis accuracy.
• Expanding the study to include more trolley stations for a comprehensive urban assessment.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

We appreciate the support and contributions from our team members and the valuable datasets provided by various sources.