Pedestrian Safety in Greater Wilshire

Henry Kaplan

Hack for LA | Data Science

The Question

Can existing datasets help us identify the parts of the Greater Wilshire neighborhood that would most benefit from new road safety features for pedestrians and cyclists?

Sources of Data

Mapillary.com

Database API

City of LA

Traffic Accidents by Date

Link: Mapillary API Documentation

​

Mapillary collects street-level imagery of roads and programmatically extracts data about the road design and features. Mapillary has given us access to their database, and Hack For LA members have built tools for working with Mapillary data that we have built on.

Link: Traffic Accidents by date

​

The City of LA has public data on traffic incidents, including pedestrian and cyclist traffic collisions, extending back to 2010. The dataset includes the police “MO” code, giving us some information about the type of each traffic collision.

Collisions with Pedestrians and Cyclists

Gathering City Data

• From the City database of traffic collisions going back to 2010, we pulled all incidents with the following police MO Codes:
• 3003 — “Veh vs Ped”
• 3008 — “Veh vs Bike”
• 3016 — “Bike vs Veh”
• 3501 — “Ped Actions”

​

• Using the City’s shapefile for neighborhood council boundaries, we matched the GPS coordinate of the collisions with a neighborhood council. (If possible. Some records fall outside LA city boundaries, some are missing location data, some fall outside the neighborhood council boundaries defined in the city shapefile.)

NC boundary source: Neighborhood Councils (Certified) | Los Angeles - Open Data Portal

The Resulting Dataset of Vehicle vs. Pedestrian & Cyclist Collisions

Collisions By Neighborhood Council

We were curious how Greater Wilshire NC compares to other neighborhood councils.

​

In both total numbers and rates adjusted for geographic size, Greater Wilshire was far from the worst. It also did not stand out when compared to neighboring district.

Incidents by Time of Day

To see if more lighting could improve safety, we charted accidents by time of day.

​

But any effect darkness may have is dwarfed by rush hour. Accidents go down as traffic decreases at night.

​

So there’s no clear indication that more lighting is needed, and lighting wouldn’t target the times and places when most accidents occur.

Mapping Safety Features

Our First Attempt

We began by pulling all records from Mapillary of signs about pedestrians, crosswalks, flashing light warnings.

Categories:

• 4543 - warning--pedestrians-crossing--g4
• 119 - regulatory--bicycles-push-button--g2
• 108 - regulatory--stop-here-on-red-or-flashing-light--g1
• 105 - regulatory-use-crosswalk--g1
• 61 - regulatory--pedestrians-push-button--g1
• 49 - regulatory-pedestrians-push-button--g2
• 11 - regulatory--turning-vehicles-yield-to-pedestrians--g1
• 1 - regulatory--stop-her-on-red-or-flashing-light--g2
• 1 - regulatory--in-street-pedestrian-crossing--g1

We found that Mapillary’s sign database was not a good proxy for pedestrian safety features.

Mapping Crosswalks

Rather than mapping pedestrian warning signs, we next mapped crosswalks.

​

Using Mapillary’s separate “features” database, we looked at physical features catergorized as pedestrian lights or crosswalks.

Mapillary categories included:

object--traffic-light—pedestrians

marking--discrete--crosswalk-zebra

construction--flat--crosswalk-plain

Comparing Collision Locations to Crossing Locations

Overlaying Safety Features

With Collision History

To look for places with more collisions and fewer safety features, we created a graphic with three overlays:

​

🟢 A green heat map of pedestrian crossing elements.

🔴 A red heat map of recorded pedestrian and cyclist collisions.

⃝ A cluster analysis of pedestrian and cyclist collisions, to sum up the total number of collisions roughly at or near each intersection.

Clearing Unhelpful Information

To make it easier to spot patterns, we whittled away data that didn’t help us.

​

• Charted collisions as clusters
• Removed clusters with fewer than 5 collisions on record
• Removed clusters outside of Greater Wilshire

The Final Map

Standout area: Wilshire Blvd between Highland and Wilton

​

• A pattern of collisions with pedestrians & cyclists at many intersections in a row.
• Relatively unfriendly to pedestrians, with busy wide streets, and relatively sparse crosswalks.
• Unlike the hotspots of collisions on La Brea, Wilshire Blvd has many intersections without crossings, and would likely benefit from more.

Takeaways

• Looking for places with few pedestrian crossings and a history of pedestrian and cyclist collisions, the stretch of Wilshire Blvd between Highland and Wilton stands out.

​

• We don’t see a need for more lighting — any effect nightfall has is dwarfed by other larger effects, like the rise and fall of traffic.

​

• Overall Greater Wilshire is not one of the hotspots of pedestrian and cyclist collisions in the City.

Possible Next Steps

• Differentiating between bicycle and pedestrian collisions
• Checking if pedestrian and cyclist hazards in the area go hand in hand, or if certain places are hotspots for one or the other.

​

• More Advanced Methods of Identifying Hotspots
• Weekdays vs weekends
• Using Friday night/Saturday morning data to see impact on non-driving shabbat temple-goers
• Comparing car-vs-car collision frequency to pedestrian/bike collision frequency (which could tell us where more lighting is needed, even as traffic falls off after dark).
• Comparing to traffic congestion data
• Comparing miles of road to number of accidents in a neighborhood

​

• Researching other effects of road design on traffic safety, such as…
• Speed limits
• Width of streets
• Number of lanes
• Medians / pedestrian islands

Questions?

Thanks to:

Contributors:

• Mapillary for donating access to their API
• Regan Hutson for arrangement with Mapillary
• Albert Ulysses for writing Hack for LA’s code for accessing Mapillary data

Henry Kaplan�Derek Plemons

Sarah Nooravi

Sophia Alice

Ryan Swan

Data Science Leads

Sophia Alice & Ryan Swan

hackforla.org

Scripting and analysis notebooks available at: https://github.com/hackforla/data-science/tree/master/Pedestrian_Data