Investigation
Bend Privacy Alliance / Axon patents series, Part 4 of 11 · September 2026
Oregon’s highest court is taking up four weeks of warrantless police surveillance outside one house. The questions it raises may matter far beyond pole cameras.
For approximately four weeks, a police camera watched a house on Baldwin Street in Salem.
Mounted on a public telephone pole, the camera could be remotely repositioned and zoomed. It captured part of the exterior of the house, the driveway, and a walkway leading toward the front door. It could not see inside the house or into the backyard.
Police were investigating burglaries involving marijuana businesses and believed David Frank Lane might have been involved. Investigators wanted to determine whether he lived at or regularly visited the Baldwin Street house. Over the course of the surveillance, officers observed Lane coming and going 15 times. Ten of those observations came from the pole camera.
When Lane left the camera’s field of view, it could not follow him, and that limitation became central to the case.
On February 19, 2026, the Oregon Court of Appeals held in State v. Lane, 347 Or App 229 (2026), A183592, that the warrantless surveillance was not a “search” under either Article I, section 9 of the Oregon Constitution or the Fourth Amendment. On June 25, 2026, the Oregon Supreme Court allowed review. The case is now before the court as S072813.
The case arrives at an important moment.
Police surveillance technology is moving beyond systems that watch one particular place. Modern systems can identify vehicles, collect observations across many locations, enrich those observations with machine-generated information, combine records from different sources and, in technologies described in Axon and Fusus patents, correlate an object’s movement from camera to camera.
Here in Bend, pieces of that transition are already visible. Police use Axon Fleet 3 cameras with automated license plate recognition. Bend has also operated a Fusus/Connect Bend environment for integrating public and private cameras. More significantly, the City-hosted 2023 Fusus service agreement itself described contracted capabilities including cross-camera tracking of people and vehicles, a community camera registry, integration of license plate readers “as required,” mobile live viewing, incident heat mapping, and integration of private video and other public-safety data sources.
Those contract terms do not prove Bend activated or used every available module. They do establish that cross-camera tracking and camera/data integration were part of the platform Bend contracted for, rather than merely capabilities found later in a patent.
That makes the central question in Lane much less theoretical: What happens when a camera no longer has to follow you for a surveillance system to reconstruct where you went?
Watching a place is not the same as following a person
The Court of Appeals began from a familiar principle: police generally do not conduct a constitutional search merely by observing something exposed to public view from a place where they are legally entitled to be.
Technology does not necessarily change that rule. Cameras, lenses and similar tools can sometimes enhance or preserve observations an officer could have made without them.
But Oregon courts have also recognized another category of technology: systems that allow police to obtain information substantially different from ordinary public observation or to conduct pervasive surveillance of a person’s movements.
That distinction drove the result in Lane.
The pole camera operated continuously for approximately a month, but the Court of Appeals separated the duration of the surveillance from the scope of the information it produced. The camera recorded Lane arriving at and leaving one location. It could not continue following him once he moved beyond its line of sight.
The court contrasted that limitation with earlier Oregon cases involving technologies capable of revealing movement across a broader portion of a person’s life.
In State v. Campbell, 306 Or 157, 759 P2d 1040 (1988), police placed a radio transmitter on a vehicle and used it to track the vehicle’s movements. The Oregon Supreme Court concluded that the surveillance was a search.
In State v. Hawthorne, 316 Or App 487, 504 P3d 1185 (2021), the Court of Appeals considered real-time cell-phone location information and emphasized that persistent location tracking can reveal where a person spends time and, through those patterns, potentially expose religious, political, social or professional associations.
The camera in Lane, the court concluded, did not do that. Although its surveillance was constant, the information it captured was “quite limited in nature.”
That distinction—between watching one place and following movement across many places—is becoming much harder to maintain technologically.
There is also an important factual wrinkle in Lane. Police apparently suspected that Lane might live at the Baldwin Street property, but Lane expressly told the trial court that it was not his residence, and a records check identified a different Salem address. The Court of Appeals relied on that fact when declining to give the property the heightened constitutional protection associated with a person’s home.
The “30 days” sometimes used to describe the case also needs qualification. The actual surveillance lasted approximately four weeks. Thirty days entered the record through the trial judge’s hypothetical: if an officer could lawfully sit in the same location “30 days 24/7,” the judge reasoned, using a camera to record what the officer could have seen would not itself create a constitutional search.
The Court of Appeals essentially accepted that reasoning on the facts before it. But the opinion did not dismiss the broader problem. It acknowledged a “genuine, growing concern” about the normalization of near-constant public surveillance.
Under the court’s analysis, the important question was not simply how long the camera watched, but what the government could learn from it.
Turning passing vehicles into machine-readable events
Automatic license plate readers provide a useful comparison—not because Lane is an ALPR case, but because ALPR shows how quickly the relationship between observation and tracking is changing.
At their simplest, ALPR systems capture vehicle images, identify license plates, associate those observations with times and locations and make the records searchable.
Axon’s patent portfolio describes a much more engineered process.
One group of patents addresses the difficulty of reading plates from a moving police vehicle. In US10715738B1 and the related US11030472B2, the systems can use specialized asymmetric illumination and adjust exposure according to the distance and relative speed of another vehicle.
Another branch describes using several images of the same plate rather than relying on one perfect photograph. Plate regions can be aligned, geometrically corrected and filtered so information from multiple frames improves the likelihood of a successful read.
A later patent describes locating a plate in lower-resolution imagery and then using higher-resolution imagery for recognition, with detection and reading potentially occurring locally on a vehicle-mounted device in less than a second.
Axon has also patented systems that monitor image blur and OCR confidence, compare recognition performance over time and automatically adjust parameters or flag maintenance problems. Its newest direct ALPR patent identified in this research, issued September 1, 2026, describes a lens with different focus regions optimized for vehicles expected at different distances from a moving camera.
Those patents are not evidence that every technique is operating on every Axon camera. They do show what the systems are being engineered to accomplish.
These are not ordinary video cameras that sometimes happen to catch a license plate. They are designed to turn passing vehicles into reliably machine-readable events.
Once that happens, the plate can become only the beginning of the record.
From one camera to a network
The broader Axon/Fusus portfolio changes the scale of the problem.
Fusus technology is built around bringing cameras and other information sources into a common real-time crime center environment, an architecture reflected in the US11368586B2 patent family. Current Axon documentation supports ALPR searches across multiple kinds of cameras and supported outside providers.
Related Fusus patents describe capabilities that extend beyond reading license plates.
One patent family describes an operator tracking a vehicle or person visible on one camera while the system identifies nearby cameras likely to see the same object next.
Another begins with an event at a known place and time, then identifies cameras that may have captured it and locates an object in the resulting video. From there, the system can estimate speed and direction, calculate where the object could plausibly travel, and select additional cameras within that predicted area.
For vehicles, the projected area can account for roads, travel direction, obstacles and estimated speed. The same patent also describes reasoning backward through time to identify cameras that may have captured where an object came from.
A readable plate is not necessarily required. The patent discusses comparing visual characteristics and machine-generated feature representations across different videos.
These patent disclosures matter, but Bend’s own records add a second evidentiary layer. The City-hosted 2023 Fusus service agreement describes cross-camera tracking of people and vehicles among the contracted platform functions, alongside integration of private video, license plate readers, CAD, automatic vehicle location, drone feeds, covert-camera feeds and body cameras.
That is stronger evidence than a patent alone. It still requires an important qualification: a contracted or available function is not the same as proof that Bend activated it, configured it for a particular use, or actually used it to track someone.
The distinction at the heart of Lane nevertheless becomes sharper.
The pole camera could not follow Lane when he left Baldwin Street. A network does not necessarily need one camera to do so.
The pieces are already here
Bend Police has used ALPR-enabled Axon Fleet 3 cameras in patrol vehicles since 2023. The underlying 2022 Axon quote included 65 Fleet 3 ALPR licenses, with license dates running from August 1, 2023 through July 31, 2028. Local reporting has described more than 70 Bend cruisers as operating with ALPR-equipped Fleet 3 systems, while approximately a dozen mobile ALPR cameras could be active at a time.
Bend’s surveillance history also includes a recent retreat. The city shut down its four stationary Flock Safety ALPR cameras on January 7, 2026 after public controversy surrounding the system. Officials later considered whether stationary Axon ALPR could replace them, but no later primary record reviewed for this project has established that such a purchase or deployment occurred.
In March 2023, Bend approved a three-year Fusus agreement for up to $230,000. The issue summary, quote and service agreement contemplated up to 150 data points or public/private video feeds, CAD integration, a community camera registry, cloud storage, mobile live viewing, an AI appliance and integration of license plate readers “as required.”
The same City-hosted agreement described a broader platform that included cross-camera tracking of people and vehicles, incident heat mapping, panic alerts with geolocation and automatic docking of nearby cameras, 911 caller video links and integration of public and private video sources. Bend-specific Fusus terms provide additional local contractual context. The agreement therefore provides stronger local evidence than the patent portfolio alone: cross-camera tracking was part of the platform Bend contracted for, even though the public record reviewed here does not establish which modules were actually activated or used.
Those contract terms establish availability, but they are silent on whether Bend enabled every function.
The network has also moved into documented local operation through Connect Bend. The Bend Bulletin reported that on March 4, 2026 the program had 618 registered cameras and 174 integrated cameras. An August 9, 2026 observation of the live Connect Bend site displayed 764 registered cameras and 322 integrated cameras.
Registration and integration are materially different. Registration identifies where a camera is located so police may request footage after an incident. Integration can provide conditional real-time access under the participating owner’s settings, as described in the Connect Bend Data Share and License Agreement.
Bend’s Connect Bend Privacy FAQ goes further in describing the platform. It says Fusus can use AI to search user-provided video for weapons and vehicles of interest while excluding facial recognition. The same FAQ describes conditional real-time streaming from integrated cameras and a virtual panic-button function capable of activating authorized camera streams.
Those are City/vendor-published descriptions of the system, not independent confirmation of every Bend setting or use.
They nevertheless make the local surveillance environment more concrete: mobile ALPR cameras, hundreds of registered or integrated private cameras, a contracted Fusus platform expressly describing cross-camera tracking, and a common vendor ecosystem designed to bring multiple sources together. Local reporting has also described a program offering FususCORE devices to Bend retailers, extending the relevance of private-camera integration beyond residences.
The unresolved question is not whether those components exist.
It is which connections among them are actually enabled, under what rules, and with what audit trail.
When observation becomes inference
Current Axon products add another layer: machine-generated information about the vehicle itself.
The Fusus ALPR environment can work with records containing information beyond plate text, including vehicle color, make, body or vehicle type, location and source.
Axon’s July 2026 Fusus release documentation also describes a feature called VAR Inference. According to the company, the feature can enrich ALPR records when vehicle color, make or model information is absent or has low confidence.
The technical provenance of that model remains unresolved in the research. Public materials reviewed so far do not identify its architecture, developer, training provider, third-party components or precise execution location. Nor does the available evidence establish that Bend has VAR Inference enabled.
But the distinction between observation and inference matters.
A camera may directly record that a vehicle was present at a particular location and time. Software can then add conclusions about what kind of vehicle appears in the image.
Repeated observations can create something further: this vehicle appeared here, then there; these are its characteristics; these cameras may have captured it later; these records likely describe the same object.
Each underlying image may depict something visible in public, but the resulting knowledge is produced through accumulation and analysis.
That brings us back to the central problem in Lane.
The Court of Appeals distinguished a camera that continuously watched one place from technology capable of revealing a person’s movements across a broad range of daily activity.
On the facts before it, that distinction was relatively straightforward. The Baldwin Street camera could tell police when Lane appeared at the property. Once he left its field of view, it lost him.
Now apply the same framework to a network.
A patrol vehicle records a car leaving one neighborhood. A fixed ALPR sees the same plate several miles away. Another camera records it later. A privately owned camera integrated into a police network captures a similar vehicle near another location. Software helps place those observations into chronological order or determine whether they depict the same object.
No individual camera followed the vehicle, yet the combined system potentially can.
The constitutional problem is what aggregation creates
This is where the traditional analogy to an officer standing on a public street begins to strain.
A person beside the road may lawfully observe a passing vehicle. But a human observer cannot simultaneously occupy hundreds of locations. A human observer does not automatically retain every passing plate in a searchable database. Ordinary eyesight does not compare a new observation against millions of historical records, calculate confidence scores or identify which cameras an object may encounter next.
The constitutional significance of surveillance technology therefore may not always be captured by asking whether an officer standing in the same location could have seen the same thing.
There is another question:
What becomes possible after the observations are remembered together?
Oregon law already contains the beginnings of an answer.
In Campbell, the significance of the radio transmitter was movement. It allowed police to do more than observe a vehicle when officers happened to encounter it.
In Hawthorne, the concern surrounding cell-phone location information was similarly cumulative. Location tracking can expose where someone spends time and, through those patterns, potentially reveal religious, political, social or professional associations.
The same distinction helps explain why ALPR retention and aggregation matter even if each individual observation occurs on a public street.
A plate observation outside a church on one Sunday may reveal little. Repeated observations at the same church on Sunday mornings can reveal more.
The same is true of a vehicle repeatedly appearing near a political meeting, union hall, protest, lawyer’s office, medical clinic or another person’s home. The intelligence lies in the pattern, and networked surveillance makes those patterns far cheaper to construct.
That observation does not decide whether any particular police use is constitutional. But focusing only on whether each camera saw something publicly visible can miss part of what the system ultimately knows.
What Lane could—and could not—decide
The Oregon Supreme Court has not yet decided State v. Lane, and there are good reasons to be cautious about predicting its reach.
Lane expressly disclaimed that the surveilled property was his residence. The camera did not see inside the house or backyard. It captured only ten observations of him coming and going during approximately four weeks of surveillance. It could not track him once he left the property.
The Supreme Court could resolve the case narrowly around those facts.
Lane therefore should not be described as a case that will determine whether ALPR systems, real-time crime centers or networked cameras are constitutional. Those technologies are not before the Court.
But the reasoning could matter far beyond this particular pole.
Oregon’s Supreme Court now has an opportunity to further define what makes technological surveillance “pervasive”; whether duration changes the constitutional character of otherwise public observation; and how courts should distinguish tools that merely enhance ordinary observation from systems that give the government a qualitatively different picture of someone’s activities.
Those questions will only become more difficult as surveillance technology moves from recording individual events toward organizing relationships among them.
The pole camera in Lane provides an unusually clean example of an older surveillance model: one camera, one location, one field of view, approximately four weeks.
When Lane left the frame, the camera could no longer see him.
The emerging model is different.
A vehicle can be captured by a moving camera, made machine-readable, associated with a location and timestamp, supplemented with vehicle attributes, placed into a common search environment alongside observations from other sources and potentially correlated with imagery from cameras the original system never controlled.
Bend’s own records now make that distinction more than theoretical. The City contracted for a platform that expressly described cross-camera tracking of people and vehicles, operates mobile ALPR cameras, and has a community-camera program with hundreds of registered and integrated cameras.
That still falls short of showing that Bend is using every available capability.
It shows why the question should be asked.
A camera pointed at one location for four weeks has already been enough to put persistent warrantless police surveillance before Oregon’s highest court.
The systems being developed now do not necessarily need one camera to watch someone everywhere.
They can allow many cameras to remember where something was.
The harder question may eventually be not whether any one observation occurred in public, but what the government is permitted to learn when it remembers them all.