"combining cloud and edge machine learning to deliver the future of video monitoring," a...
TRANSCRIPT
Combining Cloud & Edge Machine Learning to Deliver the Future of Video Monitoring
February 22, 2017Carter Maslan & Luca de Alfaro
smart monitoring monitoring
In deluge of video, less is more
https://youtu.be/0BUWRHd_jss
Historical challenges
Problem Mitigation
Fragmented camera/NVR SDKs RTSP streams via Camio Box
Expensive cloud compute Real-time Event ranking
Limited Internet bandwidth Metadata-first pipeline
False positives Adaptive motion filters
Diverse workloads Hooks for callback extensionshttps://youtu.be/HFBrc7MyuvQ
sport utility vehicle
Camio makes video simple & useful
cameras that learn
cloud service
client firmware
Camio works with any camera on any network
Why now? First time feasible & expected
1. Compute Powerdeep learning + inexpensive cameras +cloud-connected compute capacity
2. Social Expectationtransparency & accountability +security threats + always-connected mobility
Instant answers, not hours of video
people in blue approaching east stairs
sport utility vehicle
The Camio Box and Cloud Pipelines
Removal of spurious motion
Motion Detection
Events, time compression
Box
Cloud
Indexing, storage, search
ML: important for
user?ML: scene recognition
ML: object recognition
Current
The Camio Box and Cloud Pipelines
Removal of spurious motion
Motion Detection
Events, time compression
Box
Cloud
Indexing, storage, search
ML: important for
user?ML: scene recognition
ML: object recognition
Next release
The Camio Box and Cloud Pipelines
Removal of spurious motion
Motion Detection
Events, time compression
Box
Cloud
Indexing, storage, search
ML: important for
user?
ML: scene recognition
ML: object recognition
~ June
The Camio Box and Cloud Pipelines
Removal of spurious motion
Motion Detection
Events, time compression
Box
Cloud
Indexing, storage, search
ML: important for
user?ML: scene recognition
ML: object recognition
~ Summer
The Camio Box and Cloud Pipelines
Box
Cloud
Video upload MetadataCommands:
● Interesting: Upload the video ● Not interesting: Compress video
(timelapse) then upload
● Search network for new cameras● Firmware updates● Debugging● …
Camio Box Architecture
● Uploads video + metadata
● Modular, Cloud / Box compatible components
● General-purpose task and upload managers
● Validated firmware updates
● Remote debugging and logging infrastructure
Metadata-only: Store video locally, search it from anywhere
Box
Camio Cloud
VideoStorage
MetadataLocal network
Metadata-only: Store video locally, search it from anywhere
Box
Camio Cloud
VideoStorage
MetadataLocal network
“people approaching entry”
proxyserver
Metadata-only: Store video locally, search it from anywhere
Box
Camio Cloud
VideoStorage
MetadataLocal network
30 kbps / camera max
2000 kbps / camera always
Camio hardware
VM
Box ● Up to 3 cameras● ARM A53● Quad-core● 2 GHz● 2 GB RAM● Gbit ethernet
Box Pro ● Up to 16 cameras● Intel i5-5200U● Dual core (8 equiv)● 2.2 GHz● 8 GB RAM● Gbit ethernet
Box Virtual ● Linux VM● Deploy on local
hardware● Grow with the need● Download and go!
Current hardware platform challengesProblem Wish
Painful SDKs, toolchains General programmability, package repositories
ML expensive on cloud, slow on Camio Box
SoC support for ML (e.g., TensorFlow)
Limited internet bandwidth Dual-resolution video streams
Limited local processing power Efficient video decoding
OTA firmware upgrades Simple atomic OTA updates
video monitoring just got simple
Making real-time information from real-world events useful and accessible to people, apps, and services via intelligent video monitoring of any video source.