A query engine for zero-streaming cameras

Xu, Mengwei; Xu, Tiantu; Liu, Yunxin; Liu, Xuanzhe; Huang, Gang; Lin, Felix Xiaozhu

doi:10.1145/3372224.3417317

Cited by 5 publications

(6 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…NN counters and ground truth counts Table 2 lists the NN counters used in experiments. Following prior work [56,60,98], we treat as the ground truth the counts returned by the golden NN counter (YOLOv3). Alternative designs we compared:…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Approximate Query Service on Autonomous IoT Cameras

Xu,

Zhang,

Liu

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

Surveillance IoT cameras are becoming autonomous: they operate on batteries without connecting to power wires, and run analytics without help from external compute resources. Such autonomous paradigm significantly frees video analytics from privacy concern and deployment constraint. However, they are constrained by limited energy and on-device compute resources.We present a system AutCam to support counting query, an important query class, on autonomous cameras. The system is built atop three key ideas. First, AutCam identifies and utilizes the rich trade-off between sample quality and sample quantity on a given time window. Second, AutCam introduces a split-phase sampling planner to judiciously allocate energy across multiple time windows, and also balance the capture and process energy. Third, AutCam constructs a unified bounded error by integrating the errors from both sampling and inaccurate neural networks. Tested on over 1,000hour videos, our prototype of AutCam can provide good estimation of ground-truth object counts with bounded, narrow error under typical solar-harvested energy, and achieve up to 10X energy saving over baselines. How does AutCam operate?To answer a counting query, Aut-Cam processes sparse samples of image frames, as shown in Figure 1. Over the course of each aggregation window, AutCam periodically wakes up and captures frames. Every N windows (i.e. a planning horizon), AutCam materializes per window aggregated counts: it counts objects on captured frames by running neural nets (called NN object counters) on individual frames; based on per frame object counts, AutCam derives per window aggregated counts for all these windows. The notion of planning horizon is common in energyconstrained systems [52]; it gives the systems essential headroom for planning energy use over long timespans.

show abstract

Section: Methodsmentioning

confidence: 99%

“…The design goal of AutCam is to best use limited energy for answering queries with high confidence, i.e., with both bounded and low errors. This problem is never addressed in existing video analytics systems [56,59,60,98,99,103] to our knowledge.…”

Section: The Autcam Overviewmentioning

confidence: 99%

Approximate Query Service on Autonomous IoT Cameras

Xu,

Zhang,

Liu

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…For instance, trustlets managing user credentials or email contents do not need high storage throughput; trustlets for surveillance can be served with median to low frame rates and resolutions (e.g. 720P at 1FPS [25,58]).…”

Section: Example Trustlets Of Secure Iomentioning

confidence: 99%

“…Figure 7 shows the results. The per-frame latencies of driverlet range from 2.1s (720p) to 3.6s (1440p) which are usable to many surveillance applications that periodically sample images [58]. The per-frame latency decreases with the number of frames in a burst, because for each burst the driverlet pays a fixed cost to initialize the camera and the media accelerator.…”

Section: Correctness and Security Analysismentioning

confidence: 99%

Minimum Viable Device Drivers for ARM TrustZone

Guo,

Lin

2021

Preprint

Self Cite

View full text Add to dashboard Cite

While TrustZone can isolate IO hardware, it lacks drivers for modern IO devices. Rather than porting drivers, we propose a novel approach to deriving minimum viable drivers: developers exercise a full driver and record the driver/device interactions; the processed recordings, dubbed driverlets, are replayed in the TEE at run time to access IO devices. Driverlets address two key challenges: correctness and expressiveness, for which they build on a key construct called interaction template. The interaction template ensures faithful reproduction of recorded IO jobs (albeit on new IO data); it accepts dynamic input values; it tolerates nondeterministic device behaviors. We demonstrate driverlets on a series of sophisticated devices, making them accessible to Trust-Zone for the first time to our knowledge. Our experiments show that driverlets are secure, easy to build, and incur acceptable overhead (1.4×-2.7× compared to native drivers). Driverlets fill a critical gap in the TrustZone TEE, realizing its long-promised vision of secure IO.

show abstract

“…• Avoiding unnecessary resource provisioning. Although cameras can generate video streams 24 × 7, provisioning a dedicated cluster to analyze all of them is unnecessary since the events of interest can be rare and only last for a short period (such as flame or smoke) [13,28,52,56]. Thus, traditional resource provisioning strategies, such as configuring a cluster according to the peak utilization or one-time offline profiling, lead to low resource efficiency.…”

Section: Introductionmentioning

confidence: 99%

Towards cloud-edge collaborative online video analytics with fine-grained serverless pipelines

Zhang

Wang

Zhu

et al. 2021

Proceedings of the 12th ACM Multimedia Systems Conference

View full text Add to dashboard Cite

This paper proposes CEVAS, a cloud-edge collaborative video analytics system, which optimizes the workload between edge and cloud by taking into account the characteristics of the video streams. In the system, the optimal partition points of video analytics pipelines are dynamically selected.The key strength of the paper is to use a serverless infrastructure paradigm for online video analysis and to orchestrate the cloud and edge resources based on the forecasts about the edge resource demand and cloud cost for future time slots. As a result of these forecasts, the partition point is defined dynamically by considering the video content. CEVAS defines the optimal point as the one that provides minimum edge-cloud data transfer overhead and cloud expenditure. It has been shown that, selecting the partition point dynamically, provides higher performance in terms of transferred data and cloud expenditure, when compared to a system which uses fixed partition point. The paper also shows how CEVAS can be realised and the evaluation shows the feasibility of the proposed system. The current version of CEVAS implements object detection in a serverless function. The system can be further improved by considering the case where the inputs are video segments. The authors provided some insights on this issue and discussed potential solutions as future work.

show abstract

A query engine for zero-streaming cameras

Cited by 5 publications

References 38 publications

Approximate Query Service on Autonomous IoT Cameras

Approximate Query Service on Autonomous IoT Cameras

Minimum Viable Device Drivers for ARM TrustZone

Towards cloud-edge collaborative online video analytics with fine-grained serverless pipelines

Contact Info

Product

Resources

About