City spends $99,000 on AI to count 34,636 stray dogs

The City of San Antonio has approved a $99,000 contract for Southwest Research Institute to count loose dogs using artificial intelligence. This new method aims to verify a 2019 estimate that placed the number of unrestrained dogs in the city at 34,636. The project is designed to provide Animal Care Services with more precise data to direct resources effectively.

The technology relies on cell phones mounted in windows across 32 specific areas of the city. These devices record five-second video clips that an application analyzes in real time. The system uses computer vision to identify dogs, cats, cyclists, and cars within the footage. It is tuned to be permissive, storing any clip it believes contains a dog for further review.

Mike Koets, a staff engineer leading the project, explained that the system uses neural networks trained on a large set of images. He noted that the technology is not the large, data-hungry AI often in the news but rather a simpler, earlier form of computer vision. If the software identifies a potential dog, the video is saved for additional analysis on a desktop. Human reviewers also examine the clips to ensure accuracy.

Koets said the automated system has proven more reliable than manual counting in some tests. During development, researchers compared the AI's output with human observation. In several cases, the software flagged dogs that human counters had missed. Upon reviewing the saved video, the researchers confirmed the animals were present in the frame.

This study follows a 2019 survey that relied on volunteers to count dogs in the same 32 locations. That initial effort observed 139 dogs, which led to the citywide estimate of 34,636. Animal Care Services Director Jon Gary stated that the new study provides another tool to understand where unrestrained dogs appear. He added that the data will help promote responsible pet care and support the health and safety of people and pets.

The city previously attempted to update the 2019 data through a partnership with the University of Texas at San Antonio and St. Mary's University. That study began in September 2024 but faced issues with data collection. An Animal Care Services spokeswoman said the 2024 effort lacked statistical reliability due to gaps in data and varying volunteer experience levels. Some observation sheets were returned blank from areas known for high dog populations.

The 2019 study used animal welfare volunteers and staff, while the 2024 attempt relied mostly on student volunteers without staff participation. The spokeswoman suggested that the original volunteers had a better understanding of where to look for dogs that hide or stay within specific neighborhoods. The new SwRI project avoids these manual recording limitations by using machine vision.

Southwest Research Institute plans to collect most of its data in November. This timing allows for cooler weather when dogs may be more active and ensures Halloween decorations have been removed. The technology could be thrown off by such visual clutter. Animal Care Services expects to have the results of the study by the end of the year.