CoChat
Sign Up
Cited by 71IoT and Edge/Fog ComputingIoT Networks and ProtocolsEnergy Efficient Wireless Sensor Networks

Mez: An Adaptive Messaging System for Latency-Sensitive Multi-Camera Machine Vision at the IoT Edge

Open access
George, Anjus, Ravindran, Arun, Mendieta, Matias, Tabkhi, Hamed·2021·IEEE Access
DOI 10.1109/access.2021.3055775
Open this paper in CoChat — free accountOpen PDF

Abstract

Mez is a novel publish-subscribe messaging system for latency sensitive multi-camera machine vision applications at the IoT Edge. The unlicensed wireless communication in IoT Edge systems are characterized by large latency variations due to intermittent channel interference. To achieve user specified latency in the presence of wireless channel interference, Mez takes advantage of the ability of machine vision applications to temporarily tolerate lower quality video frames if overall application accuracy is not too adversely affected. Control knobs that involve lossy image transformation techniques that modify the frame size, and thereby the video frame transfer latency, are identified. Mez implements a network latency feedback controller that adapts to channel conditions by dynamically adjusting the video frame quality using the image transformation control knobs, so as to simultaneously satisfy latency and application accuracy requirements. Additionally, Mez uses an application domain specific design of the storage layer to provide low latency operations. Experimental evaluation on an IoT Edge testbed with a pedestrian detection machine vision application indicates that Mez is able to tolerate latency variations of up to 10x with a worst-case reduction of 4.2% of the application accuracy F1 score metric. The performance of Mez is also experimentally evaluated against state-of-the-art low latency NATS messaging system.

AI snapshot

Written from the paper’s full text.

TL;DR

Mez is an adaptive publish-subscribe messaging system for latency-sensitive multi-camera machine vision at the IoT Edge that trades temporary video-frame quality reduction for lower wireless transfer latency. On an IoT Edge testbed with OpenPose pedestrian detection, Mez tolerated latency variations up to 10x with a worst-case F1 accuracy reduction of 4.2%.

Key findings

  • Mez achieved target latency under 100 ms with application accuracy above 95% for complex-scene video frames, where uncontrolled median latency was 260 ms for JAAD and 650 ms for DukeMTMC; settling time was less than one second.
  • With IoT camera nodes scaled from 1 to 5, Mez maintained per-frame pub-sub latency under 100 ms, while NATS showed a super-linear latency increase; Mez’s worst-case accuracy reduction was 3.3% for JAAD and less than 4.2% for DukeMTMC.
  • In end-to-end latency breakdown with 5 IoT camera nodes and complex JAAD frames, network latency dominated at 96.2% for NATS and 65.7% for Mez, with Mez’s controller overhead next at 20.5%.
  • Mez tolerated latency variations of up to 10x with a worst-case reduction of 4.2% in the application accuracy F1 score metric.

Methods

  • Testbed: five IoT camera nodes on 8-core ARMv8.2 Nvidia Jetson AGX Xavier boards, one Nvidia Titan V GPU Edge server, NETGEAR Nighthawk XR700 802.11ac (5 GHz) WiFi access point, nodes at 6 m, PTP time synchronization; video datasets JAAD and DukeMTMC.
  • Application and metric: OpenPose multi-person 2D pose detection for pedestrian detection, evaluated with F1 score at IoU threshold 0.5; latency measured as 95th percentile at 5 fps.
  • Mez design/evaluation: five lossy image transformation “tuning knobs” (resolution, colorspace, blurring, artifact removal, frame differencing), a network latency controller using PI control plus lookup tables, and an in-memory append-only circular log; evaluation compared Mez against NATS for IoT-node and subscriber scaling.

Limitations

  • Stated: Mez does not support replicated physical nodes; if an IoT node fails its camera frames become unavailable, and if the Edge server fails the entire system becomes unavailable. It also has no inherent broker restart mechanism and relies on external services such as Kubernetes.
  • Stated: security work such as RBAC and container-image authenticity verification is left to future work; only the Xavier CPU was used, not its GPU; and the paper focuses on network latency, leaving joint compute/network latency optimization to future work.
  • Evident: evaluation is limited to one machine vision application (OpenPose pedestrian detection), two datasets, five IoT camera nodes, and 802.11ac WiFi; the controller also assumes prior characterization of latency versus frame size and accuracy versus knob settings for the target deployment/application.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/access.2021.3055775

Cites

35 of 35 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

4 of 18 cited sources checked · 0 supported · 0 flagged · accuracy not yet measured

  • Findingcitation

    Blurring the video frames can reduce the video frame size by as much as 46%.

    “Blurring the video frames can reduce the video frame size by as much as 46%.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [23] source

      Section 2.2 describes capturing and streaming video frames, and Section 3.3 discusses frame compression and transmission latency. Neither passage reports blurring frames or a 46% reduction in frame size; searches for “blurring” and “46%” found no supporting passage in our copy. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
  • Findingcitation

    Removing artifact information can reduce the video frame size by as much as 98%.

    “Removing artifact information can reduce the video frame size by as much as 98%.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [23] source

      The Gabriel Platform section (page unavailable) describes streaming a subset of image frames in MJPEG format, and §3.3 (page unavailable) describes compressing frames before transmission. Neither reports removing artifact information or a 98% reduction in frame size; searches for those terms and “98%” found no supporting passage in our copy. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
  • Findingcitation

    Colorspace modifications can reduce the video frame size by as much as 62%.

    “Our choice of color space modifications can reduce the video frame size by as much as 62%.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [23] source

      Section 2.2 describes streaming image frames in MJPEG format, and Section 3.3 discusses compressing frames and transmitting them. Neither passage, nor the additional passages searched for “62%” and color-space modifications, reports a 62% reduction in frame size from changing color space. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
  • Findingcitation

    Modifying resolution can reduce the video frame size by as much as 84%.

    “Modifying resolution can reduce the video frame size by as much as 84%.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [23] source

      Section 2.2 describes capturing 640×360 frames, and section 3.5 discusses compression and transmission differences across devices. Neither reports changing resolution or an 84% reduction in frame size (searched: resolution, frame size, 84%). (LLM, conf=0.70)

      Experimental — accuracy not yet measured
  • Findingown data

    In experimental evaluation on an IoT Edge testbed with a pedestrian detection machine vision application, Mez tolerates latency variations of up to 10x with a worst-case reduction of 4.2% of the application accuracy F1 score metric.

    “Experimental evaluation on an IoT Edge testbed with a pedestrian detection machine vision application indicates that Mez is able to tolerate latency variations of up to 10x with a worst-case reduction of 4.2% of the application accuracy F1 score metric.”
    • Also stated in: § FIGURE 1. · hedged
      “Experimental results indicate that for pedestrian detection application with the OpenPose multiperson 2D pose detection benchmark [14]- [17], Mez is able to achieve the target latency in the presence of up to 10x increase in channel interference with an application accuracy degradation of at most 4.2%.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [14]- [17]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. 14
  • May combine two claims. 20
  • The quote only partly states this claim. 3

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    For the DukeMTMC frame, the detection boxes in the modified video frame are thinner than those in the unmodified video frame.

    “For the DukeMTMC frame, the detection boxes in the modified video frame are thinner than those in the unmodified video frame.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    When the number of IoT nodes is scaled from 1 to 5, Mez maintains the pub-sub latency under 100 ms while keeping the accuracy reduction below 4.2% for the DukeMTMC dataset.

    “When the number of IoT nodes is scaled from 1 to 5, Mez is able to maintain the pub-sub latency under 100ms (Figure 14a) while maintaining the accuracy reduction less than 4.2% (Figure 14b).”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Mez shows a worst-case accuracy reduction of 3.3% for the JAAD dataset.

    “Mez shows a worst case accuracy reduction of 3.3%.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For video frames with complex scene dynamics, the ratio of FIVE Lat to ONE Lat is 5.6x for the JAAD dataset and 8.4x for the DukeMTMC dataset.

    “We note that for video frames with complex scene dynamics, the ratio FIVE Lat /ONE Lat is 5.6x for the JAAD dataset, and 8.4x for the DukeMTMC dataset.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [19] source
    • cites [20]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingown data

    Video frame modification by the network latency controller causes an accuracy loss of 3.3% for JAAD and 4.2% for DukeMTMC complex scene dynamics frames.

    “Pedestrian detections for the unmodified video frames (green), and after video frame modification (blue) by the network latency controller are shown (accuracy loss of 3.3% for JAAD and 4.2% for DukeMTMC).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The FIVE Lat at 15 fps is 1.02x higher for DukeMTMC compared to 5 fps, and at 12m is 1.06x higher compared to 6m.

    “We note that the FIVE Lat at 15 fps is 1.02x higher for DukeMTMC compared to 5 fps, and at 12m is 1.06x higher compared to 6m.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    NATS could not be evaluated on the DukeMTMC dataset because its median video frame size exceeds NATS's 1 MB message size limit.

    “Since the median video frame size for DukeMTMC is greater than 1 MB, we could not evaluate NATS due to its 1 MB message size limit.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    As the number of IoT camera nodes scales from 1 to 5, Mez maintains the pub-sub latency under 100 ms for the JAAD dataset.

    “As seen from Figure 13a, when the number of IoT nodes are increased, Mez is able to maintain the pub-sub latency under 100ms.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingnone given

    A 1 GB in-memory log at the IoT camera node holds approximately 7 minutes worth of video frames, assuming 500 kB per frame and 5 fps.

    “A 1 GB of in-memory log at the IoT camera node holds approximately 7 minutes worth of video frames (assuming 500 kB per frame, and 5 fps).”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Scene dynamics, frame rate, and distance of IoT camera nodes from the Edge server are less significant factors affecting latency than peer node interference.

    “Additionally, factors affecting latency include scene dynamics; frame rate, and distance of IoT camera nodes from the Edge server are less significant.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Multiple knob combinations map to the same video frame size and network latency, but can result in different application accuracy.

    “We also note that multiple knob combinations map to the same video frame size (and hence network latency).However, these knob combinations could result in different application accuracy -which we characterize in the next section.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In all cases, the controller achieves an application accuracy above 95% with a settling time of less than one second.

    “In all cases, the controller is able to achieve an application accuracy of above 95% with a settling time of less than one second.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    NATS maintains the maximum accuracy for all cases because it always sends unmodified images.

    “Since NATS always sends unmodified images, it maintains the maximum accuracy for all cases.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For Mez, network latency is the dominant component of pub-sub latency at 65.7%, with controller overhead next highest at 20.5%.

    “For Mez, the network latency is the dominant component at 65.7%, with the controller overhead being the next highest at 20.5%.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Both Mez and NATS scale well as the number of subscribers increases from 1 to 8, with minimal degradation in latency.

    “Both Mez and NATS scale well as the number of subscribers are increased from 1 to 8 with minimal degradation in latency.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    About half the controller processing time is due to video frame modification, with video frame copying between the logs at IoT camera nodes accounting for the remaining time.

    “About half the controller processing time is due to the video frame modification, with the video frame copying between the logs at IoT camera nodes accounting for the remaining time.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Machine vision applications can tolerate errors in video frame data, which allows the use of simpler transport protocols such as UDP that do not support re-transmissions.

    “Also, in practice we observe that the ability of machine vision applications to tolerate errors in the video frame data, allows the use of simpler transport protocols (UDP) that does not support re-transmissions.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The Edge nodes and the Edge server were synchronized using the PTP synchronization protocol before the start of the measurements.

    “The Edge nodes and the Edge server are synchronized using PTP synchronization protocol [22] before the start of the measurements.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [22]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The IoT camera nodes are time synchronized to the Edge server before starting the network latency measurements using the PTP network level time synchronization protocol capable of microsecond accuracy.

    “The IoT camera nodes are time synchronized to the Edge server before starting the network latency measurements using the PTP network level time synchronization protocol capable of microsecond accuracy [22].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [22]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    The lack of a need to support update and delete operations, and the sorted video frames in the log, simplify the design of the log and prevent memory fragmentation.

    “The lack of a need to support update and delete operations, and the sorted (by timestamps) video frames in the log, simplify the design of the log and prevent memory fragmentation.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Machine vision applications at the IoT Edge do not need to support delete operations.

  • May combine two claims.

    Interpretationreasoning

    Localized processing of video streams at the Edge potentially allows for low latency operation, overcomes bandwidth limitations by reducing the size of data sent to the Cloud, and helps address privacy and legal concerns by eliminating the need to share raw video frames with Cloud vendors.

    “The localized processing of video streams at the Edge potentially allows for low latency operation, overcomes bandwidth limitations by reducing the size of the data that needs to be sent to the Cloud, and helps addresses privacy and legal concerns by eliminating the need to share raw video frames with Cloud vendors.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Network latency is dynamic due to scene changes and the intermittent nature of external interference.

    “to scene changes (simple to complex), and the intermittent nature of external interference.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationreasoning

    The on-demand transfer of video frames reduces channel interference by limiting unneeded transmission in the wireless channel.

    “The ondemand transfer of video frames reduces channel interference by limiting unneeded transmission in the wireless channel.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisereasoning

    Existing messaging systems are built for the Cloud, where machines communicate over low latency wired networks, and do not provide mechanisms to guarantee latency when operating in wireless channels with large latency variations.

    “However, the existing messaging systems are built for the Cloud, where machines communicate over low latency wired networks (Gigabit Ethernet, Infiniband), and as such do not provide mechanisms to guarantee latency when operating in wireless channels with large latency variations.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [10]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [11]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [12] source

  • May combine two claims.

    Interpretationnone given

    The latency controllers on each IoT camera node operate independently of one another, and the lack of centralized control allows the scaling of the IoT camera nodes.

    “Note that the latency controllers on each IoT camera node operate independently of one another.The lack of centralized control allows the scaling of the IoT camera nodes.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The design of Mez storage is targeted to provide low latency read-write operations.

    “The design of Mez storage is targeted to provide low latency read-write operations.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The reduced transmission also serves as a power saving opportunity at the IoT camera node.

    “Additionally, the reduced transmission also serves as a power saving opportunity at the IoT camera node.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The persistent storage is only used to reconstruct the in-memory log during reboot after node failure.

    “The persistent storage is only used to reconstruct the in-memory log during reboot after node failure (see Section IV-D).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The impact of lower sized video frames on machine vision task accuracy is dependent on the particular machine vision application.

    “In general, the impact is dependent on the particular machine vision application.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The in-memory logs are persisted on durable storage on both the IoT camera nodes and the Edge server, but all requests are served from the in-memory log to minimize storage latency.

    “The in-memory logs are persisted on durable storage (SSD/disk) on both the IoT camera nodes, and the Edge server.However, to minimize storage latency, all requests are served from the in-memory log.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: For the short time duration the video frames are held in DRAM, the data is safe from illegal access due to the volatile nature of the memory.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Due to the possible physical insecurity of the hardware, the video frames stored on disk are encrypted at rest.

    “Due to the possible physical insecurity of the hardware, the video frames stored on disk are encrypted at rest.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    Encryption/decryption and disk accesses are relatively long latency operations, motivating the avoidance of disk access in the read/write critical path.

    “The encryption/decryption, and disk accesses are relatively long latency operations, motivating the avoidance of disk access in the read/write critical path.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    NATS is a general purpose messaging system and does not provide latency guarantees.

    “However, unlike Mez, NATS is a general purpose messaging system, and does not provide latency guarantees.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [11]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationnone given

    The current version of Mez does not support replicated physical nodes, neither IoT camera nodes nor Edge server.

    “Regarding service availability, the current version of Mez does not support replicated physical nodes (neither IoT camera nodes, nor Edge sever).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    A low quality H.264-encoded 1080p stream is insufficient to perform accurate analysis for vision analytic applications such as traffic monitoring and pedestrian tracking.

    “However, Canel et al. [26] suggest that a low quality H.264-encoded 1080p (1920 × 1080 pixels) stream is insufficient to perform accurate analysis for vision analytic applications such as traffic monitoring and pedestrian tracking.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Canel et al. [26]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

35 of 35

  • The authors state this as an assumption.

    Network latency for different video frame sizes is available from prior characterization of the video frames in the targeted deployment environment.

    “The network latency for different video frame sizes (see Section II-C) are assumed to be available from prior characterization of the video frames in the targeted deployment environment.”

  • The authors state this as an assumption.

    Application accuracy for different tuning knob settings is available for the targeted application through prior characterization.

    “The application accuracy for different tuning knob settings is also assumed available for the targeted application through prior characterization.”

  • The authors state this as an assumption.

    Real-time machine vision applications consume the data within the time frame held by the 1 GB in-memory log.

    “The real-time machine vision applications are assumed to consume the data within this time frame.”

  • The authors state this as an assumption.

    For the short time duration the video frames are held in DRAM, the data is safe from illegal access due to the volatile nature of the memory.

    “For the short time duration the video frames are held in the DRAM, the data is assumed to be safe from illegal access due to the volatile nature of the memory.”

  • The paper relies on this without support.

    Machine vision applications at the IoT Edge do not need to support delete operations.

    “Mez uses an in-memory log based storage that exploits application domain specific characteristics such as the lack of need to support delete operations, to implement a simple low latency storage.”

  • The paper relies on this without support.

    Computational resources are constrained at the Edge due to power, space, and cost considerations.

    “However, we note that computational resources are constrained at the Edge due to power, space, and cost considerations.”

  • The paper relies on this without support.

    The lower layers of the network (TCP) provide accurate replication of the log between the IoT camera node and the Edge server.

    “Instead, we rely on the lower layers of the network (TCP) for accurate replication.”

  • The paper relies on this without support.

    Many machine vision applications support the approximate computing paradigm, where useful enough results can be obtained despite reduced quality input.

    “Many machine vision applications support the approximate computing paradigm, where useful enough results can be obtained despite reduced quality input.”
8 of 8

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Evaluation describes all latencies as 95th-percentile measurements but reports medians

    Minor

    The evaluation states that all latencies are measured at the 95th percentile, then describes the controller results using median latency. Those are different summary statistics.

  • DukeMTMC NATS evaluation is both ruled out and reported

    Minor

    The paper says NATS could not be evaluated on DukeMTMC because of its message-size limit, but the Figure 15 caption reports NATS subscriber-scaling results for DukeMTMC.

2 of 2

Cites

Flagged first, then by citations

3 of the 38 references couldn’t be matched to a known paper.

Research works printed but not linked: 10

  1. vermesan, 2011, Internet of Things Global Technological and Societal Trends
  2. kreps, 2011, Kafka: A distributed messaging system for log processing, Proc of Workshop on Networking Meets Databases (NetDB), 1
  3. foundation, 2020, Nats Documentation
  4. software, 2020, RabbitMQ by Pivotal
  5. corporation, 2020, Jetson AGX Xavier
  6. foundation, 2020, Jenkins
  7. inc, 2020, ET Docker
  8. foundation, 2020, gRPC
  9. vision, 2020, OpenCV documentation
  10. foundation, 2020, Kubernetes

Cited by

Sorted by citations

Related

Authors
George, A., Ravindran, A., Mendieta, M., & Tabkhi, H.
Affiliations
University of North Carolina at Charlotte (US)
ORCID
Venue
IEEE Access
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2021
Volume
9
Pages
21457-21473
ISSN
2169-3536 (ISSN-L)
DOI
10.1109/access.2021.3055775
OpenAlex
W3127153630
Semantic Scholar
f3cfcc3c14d7da1ec0deb4873d149631ead8b2f7
Cited by
71
References
38
Citations by year
2026: 1 · 2025: 10 · 2024: 13 · 2023: 36 · 2022: 11
FWCI
6.96
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Access (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Science Foundation (1831795)
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › IoT Networks and Protocols
  • Physical Sciences › Computer Science › Computer Networks and Communications › Energy Efficient Wireless Sensor Networks
Keywords
publish/subscribe, IoT edge computing, pedestrian detection

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review