Event-Based Vision Sensor AI Patents: Who Leads, Where the Gaps Are 2026
- Filing activity peaked in 2023 at 15 families then flattened rather than continued climbing, with the 2022 midpoint at 10 — a plateau, not a slowdown once publication lag is accounted for.
- China receives more than half of all filings 33 of 62 families route through the Chinese patent office, well ahead of the United States at 14 and every other jurisdiction combined.
- Claim density concentrates in AI computing and recognition G06N and G06V subclasses cover the bulk of families, while control-integration and sport-specific applications remain thinly claimed.
A small, concentrated field built on two methods
Event-based vision sensors report pixel-level brightness changes asynchronously rather than full frames, and pairing that output with spiking neural networks or other event-native learning methods is the specific combination this landscape tracks. The corpus is small — 62 families — which means the field is still early enough that a handful of filings, rather than broad trends, define the claim boundaries. Two method clusters dominate: converting the event stream into synchronous frames for training, and processing events natively through a spiking architecture.
Filing activity is concentrated technically in AI-computing and recognition claims rather than sensor hardware, and concentrated geographically in China. Both patterns narrow where a freedom-to-operate search needs to focus, and both leave the control-integration and application-specific branches comparatively open.
Filing trend and technology composition
Sixty-two families sit inside this search, spanning eight IPC subclasses and six receiving offices. The trend and the classification split both point to where claim density has actually built up, not where the technology is merely discussed.
Filings rose then flattened
Annual filings climbed from zero in 2017 to a peak of 15 in 2023, with 2022 sitting at the midpoint of 10. The flat-to-declining trend after the peak, combined with the roughly 18-month publication lag, means 2025 and 2026 figures will read low regardless of true filing activity.
AI computing and image recognition dominate
G06N (AI computing models) and G06V (image/video recognition) account for the largest shares of the 62 families, with G06T image-processing claims close behind. Control-oriented classes (G05D) and sport-specific applications (A63B) are present but thin, marking the edges of where filers have staked claims versus where the space is still open.
Shares are the percentage of the 62 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Event-Based Vision Sensor AI and Machine Learning with Eureka
This page is one run against one query. Ask Eureka your own question about event-based vision sensor ai and machine learning and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative filing
US12401885B2 — Target tracking method and system of spiking neural network based on event camera
A target tracking method and a target tracking system of a spiking neural network based on an event camera are provided. The method includes: acquiring a data stream of asynchronous events in a high dynamic scene of a target by an event camera as input data; dividing the data stream of the asynchronous events into synchronous event frames with millisecond time resolution; training a twin network based on a spiking neural network by a gradient substitution algorithm with a target image as a template image and a complete image as a searched image; and tracking the target by a trained twin network with interpolating a result of feature mapping to up-sample and obtaining the position of the target.Granted 2025-08-26 to Zhejiang Lab — illustrates the shift from raw event-stream processing to frame-synchronised twin-network training as the claimed method of choice.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20240028036A1 | Robot dynamic obstacle avoidance method based on multimodal spiking neural network | 23 |
| 2 | CN112712170A | 基于输入加权脉冲神经网络的神经形态视觉目标分类系统 | 17 |
| 3 | CN119477976A | 一种融合事件和RGB图像的SNN目标跟踪方法及系统 | 12 |
| 4 | CN116382267A | 一种基于多模态脉冲神经网络的机器人动态避障方法 | 9 |
| 5 | US20210397878A1 | System and method of gesture recognition using a reservoir based convolutional spiking neural network | 9 |
| 6 | CN111695681A | 一种高分辨率动态视觉观测方法及装置 | 9 |
| 7 | CN112597980A | 一种面向动态视觉传感器的类脑手势序列识别方法 | 8 |
| 8 | US20230410328A1 | Target tracking method and system of spiking neural network based on event camera | 7 |
| 9 | CN117314972A | 一种基于多类注意力机制的脉冲神经网络的目标跟踪方法 | 6 |
| 10 | US20230171492A1 | Autofocus imaging circuitry, autofocus imaging apparatus, and autofocus imaging method for an event camera | 6 |
Ranked by citation count within this corpus; older filings accumulate citations simply by being available longer, so treat this as an influence signal rather than a current-activity ranking.
Patent titles are shown in the language they were filed in, not translated, so that each record stays verifiable against the original filing — a translated title will not match in Eureka or in any national register. Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the filing pattern tells you
Sixty-two families is a small, specific corpus — small enough that individual filings move the picture, and specific enough that the gaps are as informative as the concentration.
Growth has flattened since the 2023 peak
The climb from zero filings in 2017 to 15 in 2023 shows the field building steadily for six years. The plateau since then, with the midpoint year 2022 at 10, suggests the core techniques — spiking network training methods and event-frame conversion — are now well-staked, pushing new filers toward narrower combinations or adjacent applications.
Filing is concentrated in one office
China accounts for more than half of all receiving-office filings, with the United States a distant second at 14 and India, WIPO and Europe each in single digits. Freedom-to-operate work that ignores Chinese prior art in this space is working from an incomplete picture.
Method claims outweigh device claims
Almost 80% of families touch the G06N AI-computing subclass, versus 5 in G05D control integration and 2 in A63B sports applications. The corpus is built around algorithms and training methods, not sensor hardware or physical control loops — a strong hint about where the unclaimed ground sits.
Robotics obstacle avoidance leads on influence
The most-cited record in the set addresses multimodal spiking-network obstacle avoidance for robots, with a closely related Chinese filing covering the same method. High citation counts here reflect early, foundational filings more than current filing volume — read them as influence markers, not activity markers.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to event-based vision sensor ai and machine learning, with the prior art for and against each one.
Who holds the claims
The 62 families split between Chinese university and institute filers, a small cluster of neuromorphic-chip startups that co-file across related entities, and a few large electronics companies. Recent-year activity across the most-cited names has gone quiet, which is consistent with the overall filing plateau rather than any single assignee pulling back.
Neuromorphic chip startups file across sibling entities
Shenzhen Speed-time Technology and Shanghai Speed-time Technology co-file as a pair, and a related Chengdu Speed-time and Ningbo Speed-time entity also appear in the assignee list — a filing pattern typical of a startup group building layered coverage around one core neuromorphic processing technique across regional subsidiaries.
University-institute pairs anchor the tracking cluster
Zhejiang University and Zhejiang Lab co-file together, with Zhejiang Lab holding the representative US12401885B2 grant on twin-network target tracking. This pairing pattern — university plus affiliated research institute — recurs across the Chinese side of this dataset more than it does among the corporate filers.
Sony is the clearest large-corporate presence
Sony Group and Sony Europe file jointly, positioning Sony as the main established electronics manufacturer active in this specific event-camera-plus-learning niche, alongside Tata Consultancy Services as the other large corporate name in the set.
| Assignee | Recent year | YoY |
|---|---|---|
| Tata Consultancy Services | 0 | — |
| Sony Group Corporation | 0 | — |
| Shenzhen Speed-time Technology Co., Ltd. | 0 | — |
| Zhejiang University | 0 | — |
| Dalian University of Technology | 0 | -100% |
| Shanghai Speed-time Technology Co., Ltd. | 0 | — |
| Brown University | 0 | — |
| Zhejiang Lab | 0 | — |
Where to take this analysis
This landscape identifies the concentration and the gaps. Turning either into a filing or freedom-to-operate decision means going deeper on the specific claims involved.
Chart the twin-network and input-weighted SNN claims
The two most-cited method clusters — twin-network tracking and input-weighted spiking classification — are where a new filing is most likely to run into blocking prior art. A full claim chart against US12401885B2 and CN112712170A is the fastest way to see exactly where the boundaries sit.
Build a claim chart in EurekaMonitor the Speed-time entity cluster
The Shenzhen, Shanghai, Chengdu and Ningbo Speed-time filings move together as one group's layered coverage. Tracking new publications from this cluster is a reasonable proxy for where the neuromorphic-chip side of this field is heading next.
Set up assignee tracking in EurekaTest the identified white space before filing
Sports-kinematics applications and direct event-to-actuator control both show thin IPC coverage relative to the AI-computing core. A targeted novelty search on either before drafting a first claim will confirm whether the gap is real or simply unindexed.
Run a novelty search in EurekaFrequently asked questions
Filing in this dataset is spread across Chinese universities and research institutes, a cluster of neuromorphic chip startups, and a small number of large electronics companies. No single assignee holds a runaway share of the 62 families tracked here; several of the most active filers, including Zhejiang University, Zhejiang Lab and the Shenzhen/Shanghai Speed-time entities, co-file together rather than filing solely under one name. China is also the dominant receiving office by a wide margin over the United States, India and PCT filings, which matters for anyone assessing where enforcement risk is concentrated.
In practice the search terms are used near-interchangeably across this corpus — filings describe the same asynchronous, pixel-level change-detection sensor under either label, sometimes within the same family. The functional distinction that matters for claim drafting is not the sensor name but how the asynchronous event stream is processed downstream, for example whether it is converted into synchronous frames before a neural network sees it, as in US12401885B2, or fed directly into a spiking network in its native asynchronous form. Search strategies should cover both terms plus 'neuromorphic vision' to avoid missing relevant families.
The IPC composition points strongly toward computational methods: G06N (AI computing) and G06V (recognition) together dominate the 62 families, while classes tied to physical control or device structure are comparatively thin. That means most claims in this set protect training algorithms, network architectures and processing pipelines rather than sensor chip design or packaging. A separate hardware-focused search on neuromorphic sensor fabrication would surface a different, largely non-overlapping set of assignees.
No. It claims a specific combination: converting an asynchronous event stream into millisecond-resolution synchronous frames, training a twin spiking neural network with a gradient substitution algorithm, and up-sampling the tracking result through feature-map interpolation. Trackers that use a different training method, a non-twin architecture, or a different temporal binning approach fall outside its literal claim scope. Anyone designing a tracker in this space should read the granted claims directly rather than relying on the abstract, since the abstract describes the method more broadly than the claims actually cover.
The filing count peaked at 15 in 2023 with the midpoint year, 2022, at 10, and the years since have not exceeded that peak. Part of this is a genuine plateau in new filing activity, but part is a reporting artefact: patent publication typically lags filing by around 18 months, so 2025 and especially 2026 figures in this dataset are undercounted and will rise as later applications publish. Treat the most recent two years as provisional rather than as evidence of a real slowdown.
Research Event-Based Vision Sensor AI and Machine Learning in depth with Eureka
Go past this page: query the whole event-based vision sensor ai and machine learning corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.