DCT
1:23-cv-03471
Hayden Ai Tech Inc v. Safe Fleet Holdings LLC
Key Events
Amended Complaint
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Hayden AI Technologies, Inc. (Delaware)
- Defendant: Safe Fleet Holdings LLC (Delaware); Safe Fleet Acquisition Corp. (Delaware); Seon Holdings Corp. (Washington); Seon Design (USA) Corp. (Washington); Rear View Safety, Inc. (Affiliate with principal place of business in New York); and named individuals.
- Plaintiff’s Counsel: White & Case LLP
- Case Identification: 1:23-cv-03471, E.D.N.Y., 02/26/25
- Venue Allegations: Plaintiff alleges venue is proper in the Eastern District of New York because Defendants conduct business and offer products for sale in the district, and maintain a regular and established place of business in the district through affiliate Rear View Safety, Inc., located in Brooklyn, New York.
- Core Dispute: Plaintiff alleges that Defendant’s "ClearLane" automated bus lane enforcement system infringes two patents related to AI-powered mobile systems for detecting traffic violations.
- Technical Context: The technology involves using cameras and artificial intelligence on moving vehicles, such as city buses, to automatically identify, document, and process traffic infractions like illegally parked cars in dedicated bus lanes.
- Key Procedural History: The filing is a Fourth Amended Complaint. The complaint includes extensive allegations of trade secret misappropriation, claiming Defendants gained improper access to Plaintiff's confidential investor presentation and engaged in "product espionage" by photographing and analyzing Plaintiff's systems installed on New York City MTA buses during a pilot program.
Case Timeline
| Date | Event |
|---|---|
| 2018 (on or before) | Defendants began discussions with New York MTA regarding an Automatic Bus Lane Enforcement (ABLE) system. |
| 2019-06-21 | Defendants submitted a proposal to the New York MTA for an ABLE system. |
| 2020-10-16 | ’919 Patent Priority Date |
| 2020-11-09 | ’014 Patent Priority Date |
| 2021-05-11 | U.S. Patent No. 11,003,919 Issues |
| 2021-11-02 | U.S. Patent No. 11,164,014 Issues |
| 2025-02-26 | Fourth Amended Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 11,003,919 - "Systems and Methods for Detecting Traffic Violations Using Mobile Detection Devices"
The Invention Explained
- Problem Addressed: The patent addresses the high rate of false positives (up to 80%) associated with prior art automated traffic enforcement solutions, which were often logic-based and not well-suited for enforcing lane violations beyond fixed intersections (Compl. ¶96; ’919 Patent, col. 1:40-44).
- The Patented Solution: The invention proposes a system of mobile "edge devices" mounted on vehicles. An edge device captures video of a potential violation at a first point in time, processes it using a deep learning model to identify the vehicle and restricted area, and transmits an evidence package to a server. A second observation from another edge device at a later time is also sent to the server, which then makes a final determination that a violation has occurred based on an elapsed time and a comparison of data from the two captures (’919 Patent, col. 2:62-67). This multi-device, time-based corroboration is central to the claimed method.
- Technical Importance: The system aims to create a more accurate and scalable method for detecting traffic violations by leveraging deep learning models on mobile platforms and using data from multiple sources to confirm infractions (Compl. ¶98).
Key Claims at a Glance
- The complaint asserts infringement of at least independent Claim 1 (Compl. ¶102).
- The essential elements of Claim 1 include:
- At a first edge device, capturing a first video of a vehicle and restricted road area at a first point in time.
- Using a processor on the first edge device to identify the vehicle, its attributes, and its license plate by applying a computer vision library and a deep learning model.
- Bounding the vehicle and restricted area in a first frame using bounding boxes.
- Detecting a first potential violation based on the overlap of the bounding boxes and transmitting data to a server.
- At a second edge device, repeating the capturing, identifying, and bounding steps at a second point in time to detect a second potential violation and transmitting that data to the server.
- At the server, determining that a traffic violation has occurred based on an elapsed time between the two points in time and a comparison of the data from the two edge devices.
U.S. Patent No. 11,164,014 - "Lane Violation Detection Using Convolutional Neural Networks"
The Invention Explained
- Problem Addressed: Similar to the ’919 Patent, this patent addresses the high false positive rates of prior art logic-based enforcement systems that could not accurately distinguish violating vehicles from other objects or normal traffic maneuvers (Compl. ¶125; ’014 Patent, col. 1:50-54).
- The Patented Solution: This invention discloses a specific AI architecture for more robust detection. The method involves using two separate convolutional neural networks (CNNs) on an edge device. A first CNN is used to detect and bound a vehicle in a video frame. A separate, multi-headed second CNN is used to detect and bound a plurality of roadway lanes, including a "lane-of-interest." A potential violation is then detected based on the overlap between the vehicle's bounding box and the lane-of-interest's polygon (’014 Patent, Abstract; col. 2:8-27).
- Technical Importance: The patent claims a specific and technically complex arrangement of AI models, separating the tasks of vehicle detection and lane detection into different neural networks to achieve a more scalable and accurate result than prior systems (Compl. ¶127).
Key Claims at a Glance
- The complaint asserts infringement of at least independent Claim 1 (Compl. ¶131).
- The essential elements of Claim 1 include:
- Cropping and resizing video frames from an edge device's video sensor.
- Bounding a vehicle detected in the frames in a vehicle bounding box, wherein the vehicle is detected and bounded using a first convolutional neural network.
- Bounding a plurality of roadway lanes in a plurality of polygons, wherein the lanes are detected and bounded using multiple heads of a multi-headed second convolutional neural network separate from the first.
- Identifying at least one of the polygons as a lane-of-interest (LOI) polygon.
- Detecting a potential traffic violation based on an overlap of at least part of the vehicle bounding box and the LOI polygon.
III. The Accused Instrumentality
Product Identification
- The accused product is the "ClearLane" automated bus lane enforcement system, also referred to as the "SF ABLE System" (Compl. ¶¶5, 74).
Functionality and Market Context
- The ClearLane system is described as being mounted on buses and relying on a "context and an ALPR camera, a purpose-built computer with intertial sensors, a GPS receiver and a cellular router" (Compl. p. 12). The system is designed to capture license plate details of vehicles in violation of bus lane restrictions, use "advanced algorithms" to determine the position and duration of the violation, and assemble an "evidence package" that is sent to a server for review and ticket processing (Compl. ¶¶62, 104, 112). A marketing graphic for the ClearLane system depicts a five-step process: capturing license plate details, identifying the vehicle, processing rules against the vehicle, forming an evidence package, and sending the package for review (Compl. p. 12). The complaint alleges Defendants market ClearLane as a direct competitor to Plaintiff's Hayden ABLE System for contracts with municipal transit agencies (Compl. ¶60).
IV. Analysis of Infringement Allegations
’919 Patent Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| capturing a first video of a vehicle and a restricted road area using one or more video image sensors of a first edge device... | The ClearLane system uses cameras mounted on a bus to capture video of vehicles that "obstruct bus lanes" (Compl. ¶104). | ¶104 | col. 2:1-4 |
| identifying, using one or more processors of the first edge device, the vehicle...by applying a plurality of functions from a computer vision library...and passing...to a deep learning model running on the first edge device | The ClearLane system uses "[a]dvanced algorithms" that run on an on-bus computer to "determine the position and location of the violating vehicle" and identify "license plate details" (Compl. ¶109). | ¶109 | col. 2:5-10 |
| bounding, using the one or more processors of the first edge device, the vehicle and the restricted road area...in a plurality of first bounding boxes | The ClearLane system employs the use of bounding boxes to identify vehicles and restricted road areas, as depicted in a marketing image showing vehicles and a bus lane outlined (Compl. ¶111). | ¶111 | col. 2:10-12 |
| detecting, at the first edge device, a first potential traffic violation...based in part on overlap of the plurality of the first bounding boxes and transmitting...data from the first edge device to the server | The system detects a potential violation and sends an "evidence package" containing vehicle attributes and positioning data to a server for "review and ticket processing" (Compl. ¶112). | ¶112 | col. 2:12-19 |
| capturing a second video of the vehicle...using one or more video image sensors of a second edge device...at a second point in time after the first point in time | The system is designed to identify violators by "matching the images and metadata captured by multiple buses," which operate at different points in time (Compl. ¶115). | ¶115 | col. 2:28-32 |
| determining, at the server, that a traffic violation has occurred based on an elapsed time between the second point in time and the first point in time and based on a comparison of the...data | The system determines the "length of time" a vehicle has been blocking a bus lane and uses a centralized server where "metadata from multiple buses can be matched and correlated" (Compl. ¶¶115, 116). | ¶¶115, 116 | col. 2:62-67 |
- Identified Points of Contention:
- Scope Questions: A central question will be whether the accused system's server-side processing meets the claim limitation of "determining...a traffic violation has occurred based on...a comparison of the first alphanumeric string with the second alphanumeric string." The complaint alleges the system matches metadata from multiple buses to identify violators (Compl. ¶115), but the specific nature of this server-side comparison will be a key factual dispute.
- Technical Questions: What is the specific nature of the "deep learning model running on the first edge device"? While the complaint alleges the use of advanced on-board algorithms (Compl. ¶109), the defendant may argue that significant processing is offloaded to the server, which could create a mismatch with the claim's requirement for certain identification and bounding steps to occur on the edge device itself.
’014 Patent Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| cropping and resizing one or more video frames of a video captured by one or more video image sensors of an edge device | The ClearLane system "employs cameras to capture the license plate details by cropping and resizing a video frame from a video device" (Compl. ¶135). | ¶135 | col. 2:9-11 |
| bounding...a vehicle detected from the one or more video frames...in a vehicle bounding box, and wherein the vehicle is detected and bounded using a first convolutional neural network | The ClearLane system uses bounding boxes to identify vehicles, as shown in a marketing image where a van is highlighted in a red box, and uses "[a]dvanced algorithms" to determine its position (Compl. ¶¶136, 137). | ¶¶136, 137 | col. 2:11-15 |
| bounding...a plurality of lanes of a roadway detected from the one or more video frames in a plurality of polygons, wherein the plurality of lanes are detected and bounded using multiple heads of a multi-headed second convolutional neural network separate from the first... | The system identifies and bounds lanes, as depicted in a marketing image showing a bus lane outlined in blue, and is alleged to use a deep learning model (Compl. ¶¶85, 137). | ¶¶85, 137 | col. 2:15-23 |
| detecting...a potential traffic violation based in part on an overlap of at least part of the vehicle bounding box and at least part of the LOI polygon. | A marketing image for the ClearLane system depicts a vehicle (outlined in a red bounding box) physically overlapping with the designated bus lane (outlined in blue), suggesting an operational basis for this detection method (Compl. ¶¶136, 139). | ¶¶136, 139 | col. 2:24-27 |
- Identified Points of Contention:
- Technical Questions: The primary point of contention will be evidentiary: does the ClearLane system's software architecture actually employ a "first convolutional neural network" for vehicles and a "separate...multi-headed second convolutional neural network" for lanes? The complaint provides evidence of the system's output (bounding boxes on vehicles and lanes) but not direct evidence of its internal AI architecture. Proving this specific, complex structure as claimed will be a significant technical challenge for the plaintiff.
- Scope Questions: What is the scope of a "multi-headed" CNN? The defendant may argue that even if its system uses a sophisticated AI model for lane detection, it does not have the specific "multi-headed" structure described and enabled by the patent's specification.
V. Key Claim Terms for Construction
The Term: "edge device" (from ’919 Patent, Claim 1)
Context and Importance: The claims require significant processing—including running a deep learning model—to occur "on the first edge device." The definition of this term is critical because if the accused on-bus hardware is merely a sensor package that streams raw video to a server for all intelligent processing, it may not meet this limitation. Practitioners may focus on this term to dispute where the inventive steps are actually performed.
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The patent specification broadly illustrates the edge device 102 as a unit that can be installed in various carrier vehicles, suggesting flexibility in its form factor (’919 Patent, Fig. 4).
- Evidence for a Narrower Interpretation: The specification details that the "processors 200" of the edge device can comprise a GPU, CPU, and a deep learning accelerator (DLA), implying a requirement for substantial, specialized on-board computing power, not just a simple camera and transmitter (’919 Patent, col. 13:11-21).
The Term: "a multi-headed second convolutional neural network separate from the first convolutional neural network" (from ’014 Patent, Claim 1)
Context and Importance: This term defines the core technical distinction of the invention. Infringement will likely depend entirely on whether the accused system's architecture can be proven to map onto this specific two-network, multi-headed structure. The defendant will likely argue its system uses a single, unified AI model or a different non-infringing architecture.
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The patent does not appear to provide an explicit definition of "separate," which may allow for an argument that logically distinct processing paths within a single, larger model could satisfy the limitation.
- Evidence for a Narrower Interpretation: Figure 7 of the patent depicts a "First Worker" performing vehicle detection and a "Second Worker" performing lane detection in parallel, with outputs being sent to a "Third Worker." This architectural diagram strongly supports an interpretation requiring two computationally distinct and separate processes or models (’014 Patent, Fig. 7).
VI. Other Allegations
- Indirect Infringement: The complaint alleges that Defendants induce infringement by providing the ClearLane system to customers, such as the New York MTA, and providing product specifications, user manuals, and technical support that encourage and assist in the infringing use of the system (Compl. ¶¶179, 193).
- Willful Infringement: The complaint alleges willfulness based on Defendants' actual knowledge of the patents since at least the filing of the original complaint, and on constructive knowledge prior to the suit due to the parties being direct competitors in the automated bus lane enforcement market (Compl. ¶¶180, 194).
VII. Analyst’s Conclusion: Key Questions for the Case
- A central issue will be one of evidentiary proof of architecture: Can Plaintiff produce evidence demonstrating that the accused ClearLane system internally operates using the specific "separate, multi-headed" convolutional neural network structure recited in Claim 1 of the ’014 Patent, or will its infringement theory rely on inferring this structure from the system's observed functionality?
- A key question of claim scope and function will concern the ’919 Patent: Does the accused system's method of correlating data from "multiple buses" on a "centralized server" meet the claim requirement of a server-side determination of a violation based on a comparison of specific data sets captured at two distinct points in time?
- A significant contextual factor will be how the extensive allegations of trade secret misappropriation and "product espionage" influence the proceedings. While separate from the patent infringement analysis itself, this narrative may be used to frame arguments regarding intent, motivation for infringement, and willfulness.