A Field Guide to Police Surveillance Technology You May Encounter

TL;DR

Modern US policing runs on seven overlapping tools: automatic license plate readers, body-worn cameras, "drone as first responder" programs, cell-site simulators (Stingrays), face-matching systems, acoustic gunshot detection, and crewed wide-area aerial surveillance. Each has a recognizable form on the ground, a published capability on paper, and a real limit you should know before you read the next breathless press release. None of these tells is unique on its own; the cluster of signals matters more than any single feature.

What "Police Surveillance" Means Now

Five years ago, the phrase mostly meant a patrol car's dashboard camera and a body mic. The category has expanded into something closer to an always-on sensor stack: cameras that read plates without aiming at them, microphones that listen for gunshots, drones launched before the first officer arrives, and software that turns any still image into a probable-identity search. This guide walks through the seven tools you are most likely to encounter, what each looks like from the outside, and what the vendor's own documentation says it can do.

Two caveats before we start. First, the data these tools generate is often retained far longer than the incident that produced it: a Brennan Center analysis of agency retention policies found Seattle PD kept ALPR reads for 90 days, San Francisco PD and the LA County Sheriff for 2 years, and Boston PD indefinitely (as of 2012).^[1] Second, "police" is not the only operator. Many of these tools are sold to private companies, malls, homeowner associations, and apartment complexes, and the data they collect is regularly shared back to law enforcement. A community camera mounted on an apartment gate and a police plate reader on a pole now feed the same queries.

Automatic License Plate Readers (ALPRs)

ALPRs are cameras built to read plates, not to watch people. A fixed pole-mounted unit, a mobile unit on a patrol car, and a portable trailer are the three forms you will see in the field. Flock Safety, the largest US vendor, sells a lineup that includes the LPR Pro, Standard LPR, Wide-Range LPR, Long-Range LPR, Flex LPR, and LPR Trailers, and the company's product page advertises that its "nation's largest LPR network" supported "1M+ cases" in 2025 and reaches "49 states."^[2]

The scale of plate collection is the part that surprises readers. According to the Brennan Center, LAPD alone has stored more than 320 million plate scans, and "99.9 percent of the 320 million images Los Angeles stores are for vehicles that were not on a hot list."^[1] Vigilant Solutions (now owned by Motorola Solutions) reported holding more than 5 billion scans, of which 1.5 billion came from law enforcement feeds.^[1] Some ALPRs can scan roughly 2,000 plates per minute.^[1]

The honest limits: read accuracy is not perfect. The Brennan Center cites a Vallejo, California study in which 37% of fixed-reader "hits" and 35% of mobile reads were misreads.^[1] And retention policy varies wildly by agency, with no federal floor: Seattle keeps reads for 90 days, Boston keeps them indefinitely, and 80 or more local departments shared ALPR data with ICE through Vigilant's LEARN network.^[1] For a fuller visual guide to identifying the hardware itself, see our field guide to spotting ALPRs.

Body-Worn Cameras

Body cameras are the one tool police are usually happy to be seen wearing. The current-generation commercial unit is more capable than the early models. Axon's Body 4 ships with a 5MP sensor, a 160-degree field of view (advertised as 36% wider than the Body 3), 128GB of onboard solid-state storage, XTS-AES 256 encryption, an IP68 waterproof rating, and a 4300 mAh battery designed for full-shift use with magnetic fast charging that delivers 20% in 30 minutes.^[3] It also integrates with Axon's real-time evidence platform for live streaming and integrates with "Axon Signal" sensors that auto-trigger recording when a Taser is deployed or a patrol car door opens.

The thing most readers miss: a body camera is a recording device whose footage is governed by the agency's retention and release policy, not by the camera itself. Officers can mute, stop, or fail-to-activate the device depending on department policy, and footage requests are processed through the same public-records machinery as any other police record. "Body cam exists" is not the same as "footage will be released to you."

Drone as First Responder (DFR)

The newest visible change on US streets is a small drone lifting off from a rooftop box when 911 rings. Drone-as-First-Responder (DFR) programs launch a camera-carrying quadcopter to a call before the first patrol unit arrives, then stream the live video to a teleoperator who decides whether to escalate, redirect, or stand down. Flock Safety now markets "Drone as First Responder" and "Drone as Automated Security (Flock Aerodrome)" alongside its plate-reader lineup.^[2]

The visible form: a small weatherproof dock mounted on a police station roof, a parking garage, or a utility pole, with a quadcopter inside that lifts off, flies to a call, and returns to charge. The hardware is generic-consumer; the differentiator is the FAA waiver for Beyond Visual Line of Sight (BVLOS) operations that lets the drone fly miles from the operator. DFR programs typically operate under specific FAA authorizations rather than standard Part 107 rules.

Honest limit: the drone is a camera platform, not an autonomous decision-maker. It does not detain, identify, or chase anyone; the teleoperator does. If you see one overhead, an officer somewhere is watching.

Cell-Site Simulators (Stingrays and Family)

Cell-site simulators are the police tool most readers have never seen, because by design they do not look like anything. The EFF's Street Level Surveillance project catalogs a long product family: Harris Corporation's StingRay, Hailstorm, ArrowHead, AmberJack, and KingFish; Boeing's Digital Receiver Technology "dirtboxes"; and products from Keyw, Octastic, Tactical Support Equipment, Berkeley Varitronics, Cogynt, X-Surveillance, Atos, Rayzone, Martone Radio Technology, Septier Communication, PKI Electronic Intelligence, Datong, Ability Computers and Software Industries, Gamma Group, Rohde & Schwarz, and Meganet Corporation.^[4]

How they work, in plain English: a cell-site simulator broadcasts a stronger signal than the legitimate cell tower next to it, so every phone in range disconnects from the real tower and connects to the simulator instead. Once a phone is connected, the simulator can identify its IMSI (the unique subscriber number), log its location, and, if it downgrades the phone to 2G/GSM, intercept "the content of unencrypted phone calls and text messages," call metadata, and some data usage.^[4] Manufacturer materials indicate the devices can be configured "to divert calls and text messages, edit messages, and even spoof the identity of a caller."^[4]

The EFF's catalog puts concrete numbers on the scale: simulators can capture up to 10,000 phones at once, disrupt communications within roughly a 500-meter radius, and have been confirmed in use by the FBI, DEA, NSA, Secret Service, ICE, US Army, Navy, Marine Corps, National Guard, US Marshals, Baltimore Police, Annapolis Police, Detroit-area ICE, San Bernardino County Sheriff, Miami-Dade Police, Milwaukee Police, and the Texas National Guard.^[4] San Bernardino County Sheriff used a device over 300 times in a little over a year; Miami-Dade purchased its first device in 2003.^[4]

The reason this matters to a non-target reader: a cell-site simulator does not just collect the suspect. It collects every phone in the area, including yours, including any protester, journalist, or bystander who happens to walk past. The EFF's documentation calls this a "general search of all cell phones within the device's radius, in violation of basic constitutional protections."^[4]

Facial Recognition

Face-matching is the system most often described in vendor marketing. Clearview AI's homepage advertises "70+ Billion Images in our law enforcement database," claims "99+% accuracy for all demographics," and markets a separate "Clearview GovCloud" product that is "FedRAMP In Process" at the High Impact Level.^[6] The company sells search access to law enforcement, federal agencies, public defenders ("JusticeClearview"), and national-security buyers.^[6]

How it is used in practice: an investigator uploads a still image (a CCTV frame, a social-media photo, a driver's-license side capture) and receives a ranked list of candidate matches drawn from the vendor's database of public-web photos. The investigator then does the human confirmation; the system returns a probabilistic score, not an identity.

The honest limits are well documented and worth saying out loud. NIST testing and outside audits have repeatedly shown that commercial face-matching accuracy varies by skin tone, gender, and age, with error rates higher for Black faces and for women. Vendor self-claims of "99+%" do not eliminate demographic disparity, and they do not survive camera angle, lighting, or partial occlusion. There is no federal US law governing police face-matching use; the rules are set state by state, agency by agency. A face-matching "hit" is a lead, not evidence; treating it as evidence has already produced wrongful arrests.

Acoustic Gunshot Detection (ShotSpotter)

ShotSpotter, now sold by SoundThinking, is a network of acoustic sensors mounted on rooftops and utility poles that listen for the pressure signature of a gunshot. The vendor's law-enforcement page claims "97% ShotSpotter's average accuracy," "< 60 sec Time for ShotSpotter to alert authorities," and that "<20% of gunfire is reported to 9-1-1."^[5] The same site advertises "+300 Agencies Trust SoundThinking Across US" and lists the product as "SOC 2 Compliant."^[5]

The visible form: small weatherproof boxes, roughly the size of a small birdhouse, mounted on rooftops, light poles, and the sides of buildings, usually in clusters of a few per square mile in covered areas. There is no camera and no flash. The sensor is an acoustic array, and the alert that pops up on a dispatcher's screen is a triangulated location, not a video feed.

The honest limit is the one the vendor does not put on its homepage: independent reviews and reporting have questioned whether the 97% accuracy figure holds up outside controlled demos, and accuracy on lower-caliber or suppressed rounds is meaningfully worse than the marketing number. Coverage is also patchy, urban-core-only in most cities; outside the sensor footprint, the system does not exist.

Wide-Area Aerial Surveillance (PSS)

The oldest still-operating form of mass police surveillance is also the least visible: a crewed Cessna circling at altitude, recording the whole city below. Persistent Surveillance Systems (PSS), an Ohio-based firm whose president and CEO Dr. Ross McNutt holds a PhD from MIT and founded the Air Force Center for Rapid Product Development at Wright-Patterson AFB, traces its lineage to "Project Angel Fire, the first wide area surveillance system," deployed in support of the Marine Corps in Fallujah and operated for four years in Afghanistan.^[7] The civilian PSS product is described on the company's own history page as a "new generation of airborne and ground-based wide area imaging systems and services," with named customers including the Philadelphia Police Department, the Baltimore Police Department, the DEA, the El Paso Intelligence Center (EPIC), Customs and Border Protection, and the government of Mexico.^[7]

What the system does: a small plane flies a racetrack pattern over a city for many hours at a time, and high-resolution cameras on the aircraft capture continuous imagery across many square miles. An analyst can then rewind any point in the recording to see every vehicle and person that moved through the frame, including where they came from and where they went. The PSS page does not publish the camera specifications, altitude, or coverage area on its public history page, so this description rests on the company's customer list and stated mission rather than on technical specs the company has chosen to release.^[7]

Honest limit: PSS-style surveillance is rare and very expensive; only a handful of US cities have run it, and several of those programs (notably Baltimore's) ended after public reporting. The technology is not new; the willingness to deploy it continuously over a US city is.

What This Means in Practice

A few rules of thumb that hold across all seven tools.

Visual identification is probabilistic. No single feature uniquely identifies a given tool. The cluster matters. Our field guide to spotting ALPRs walks through the cluster method in detail for plate readers specifically; the same logic applies to acoustic sensors (cluster of small boxes in a tight urban grid), to body cameras (cluster on a single officer), and to DFR docks (cluster of roof-mounted boxes near a station).

"Live" surveillance is the smallest part of the system. Every tool above has a backend: a database, a retention policy, a query interface, and a list of who can search it. The vendor's homepage says almost nothing about any of these. The Brennan Center's documentation of LAPD holding 320 million scans with 99.9% not on any hot list is the more honest version of the story than any product demo.^[1]

"Not facial recognition" is a real claim, and worth checking. Flock's product page carries the line "Not included. Not planned." specifically to reassure buyers that its plate-reader network does not run face-matching on top of plate reads.^[2] Other vendors do not make that promise. When a system mixes plate reads with face-matching, the data is also mixed, and the agency's retention policy becomes the only thing standing between a single traffic stop and a permanent record.

Records laws are the most useful lever. If you want to know what your local department actually deploys, a public-records request is more reliable than a vendor sales call. Departments publish their ALPR retention policies, their cell-site simulator NDAs, and their face-matching use policies when forced to; the EFF's Stingray catalog and the Brennan Center's ALPR report were both built largely on documents that were never meant to be public.^[1]^[4]

The Bottom Line

Police surveillance today is a sensor stack, not a single tool. The seven pieces above each have a published capability, a recognizable form on the ground, and a real limit that the vendor's marketing page will not mention. The cluster of features identifies the tool, the cluster of policy documents determines what it can do to you, and the cluster of records requests is how a regular reader actually finds out what is deployed locally. None of the tells is unique; the cluster is.