Introduction
This notebook analyzes one week of Seattle Police Department (SPD) Automated License Plate Reader (ALPR) activity obtained by the University of Washington Center for Human Rights (UWCHR) through public records requests. For previous UWCHR research regarding ALPR devices see the report “Who’s Watching Washington: Dangers of Automated License Plate Readers to Immigrant and Reproductive Rights in Washington State”.
License Plate Reads
Each time a license plate is detected by the ALPR device, the system records the date and time of the detection and the location of the license plate hit. This information reveals insights about how these surveillance devices are used.
The dataset contains 93490 instances where potential license plates were detected by SPD’s surveillance system from October 1, 2021 (a Friday) to October 8, 2021 (a Friday). Of the 68188 unique license plates (and other values) captured, 57237 reads, which accounts for 84% of total unique reads, were only captured once. We also found that on average, each unique license plate was captured 1.37 times.
ALPR cameras also appear to pick up other types of visual data. For instance, the data includes reads that appear like road signs (such as AHEAD, DETOUR, CLOSED, ROAD, and WARNING) and other public signage (like POLICE, TOWAWAY, CROSSWALK, and PARKING). Though these reads did not capture license plates, we did not remove them from our analysis because they still provide potentially meaningful information about the presence of ALPR readers.
Analyzing the Spatial Deployment of SPD’s ALPR System
Figure 1 provides an overview of the geographic distribution of these reads. As a note, reads are based on the geolocation of each address provided by SPD, which may not be exact.
One ALPR device is observed to travel outside of Seattle and King County in the vicinity of the cities of Everett and Snohomish.
Figure 1: Overview of All Reads
For a closer look, Figure 2, Figure 3, and Figure 4 depict zoomed-in maps that highlight areas with the highest concentration of ALPR reads through brighter colors. The most read-dense areas are located near SPD precincts. Notably, there are no reads in West Seattle and very few reads in East Capitol Hill and waterfront neighborhoods like Magnolia. This may be due to missing data, as discussed below.
Figure 2: Heat Map of ALPR Reads
License plates are more likely to be captured on camera in the vicinity of SPD precincts.
Figure 3: Heat Map of ALPR Reads in Downtown Seattle
While license plates all throughout downtown are likely to be captured by the ALPR system, we have observed that specific neighborhoods, such as East Capitol Hill, appear to go entirely un-monitored, at least during the time period covered by this dataset.
Figure 4: Heat Map of ALPR Reads near SPD Precinct
SPD’s North Precinct, marked by the white point in the map below, is located near the Northgate Transit Center. The higher frequency of reads in this area could be attributed to patrol cars traveling to and from the precinct.
Figure 5: Heat Map of ALPR Reads near University Village
High frequency of APLR reads in the vicinity of the University of Washington Seattle campus and University Village appear to reflect a device capturing plate reads from a stationary position near NE 45th Street.
ALPR Device Characteristics
The data set consists of 9 devices. Each device captured an average of 10388 license plates, with the most device capturing 21983 plates.
Reads associated with two devices were completely missing addresses, comprising 26% of all reads in the data.
Another two devices were labeled as attached to parking enforcement officer vehicles (PEO). Reads from PEO devices account for 9% of reads in the data set. It is possible that other devices are also associated with parking enforcement activities but not labeled as such.
Figure 6: Reads Per Device
| Device | Abbr. | Reads | Missing Address | Percent of Reads |
|---|---|---|---|---|
| M207541_Deleted_2021-10-21_22 50:42 | A | 21983 | 34 | 23.5137448 |
| SPD2020681MDC | B | 16433 | 189 | 17.5772810 |
| BXH9904 SPD2020627MDC | C | 14002 | 14002 | 14.9770029 |
| M207709 | D | 11631 | 59 | 12.4409028 |
| SPD2020537MDC | E | 10394 | 10394 | 11.1177666 |
| SPD2020602MDC | F | 10334 | 11 | 11.0535886 |
| PEOATG508533 | G | 8312 | 15 | 8.8907905 |
| SPD2020621MDC | H | 358 | 1 | 0.3829287 |
| PEO508459 | I | 43 | 0 | 0.0459942 |
Figure 7: Tracking the Movements of ALPR Devices
The paths of two active ALPR devices are depicted below. The maps show where the device captured license plates on camera throughout the day.
This vehicle collected plates all throughout the Seattle area across multiple days.
Meanwhile, this vehicle was active during just one day in the data set, and it only captured plates during a trip outside Seattle and King County.
ALPR System Usage By Day and Time
Contrary to our expectations, SPD devices were overall relatively inactive during the beginning of the work day and inactive during the morning rush hour commute. Figure 8 and Figure 9 show that devices collected the most data in the middle of the work day, between 12:00 pm and 6:00 pm. On the busiest day in our data set, SPD cameras captured 22773 license plates. Up to 146 plates were captured in a single minute.
Figure 8: SPD ALPR Collection by Time of Day
ALPR activity was low in the early morning and evening. Overall, ALPR activity was busiest in the middle of the day.
Figure 9: SPD ALPR Collection by Time of Day Per Day
This trend was similar throughout the week.
Figure 10: SPD ALPR Collection by Day of the Week
ALPR activity was busier on weekdays, with a lower frequency of ALPR captures occurring over the weekend.
Figure 11: SPD ALPR Collection by Day and Device
Not all ALPR devices recorded license plate reads every day of the week. Note devices “C” and “E” are missing all address data.
Analyzing SPD’s ALPR System and Census Data
Our preliminary statistical analysis suggests a weak negative correlation between plate hits and median household income, indicating a possible increase of surveillance in lower income neighborhoods. Figure 12 below depicts this relationship. We did not identify any other suggestive correlations between SPD surveillance and other demographic factors.
However, when accounting for over-dispersion, the correlation loses its statistical significance. This finding should thus be interpreted with caution. Moreover, our analysis only covers data from a single week, and a large portion of the collected readings are missing important location data. For a brief analysis, see this companion notebook.
Figure 12: ALPR Plate Reads and Median Income
