Data Collection
& Annotation for Edge and Cloud Computer Vision
How SQUAD helped a global smart-camera brand collect 300+ TB of video and sensor data,
train high-accuracy models, and migrate most computer vision workloads
from cloud to edge.
300+ TB
of video and sensor data collected for edge and cloud algorithms
CV algorithms moved to edge,
cutting redundant cloud usage
Faster, more reliable alerts
and quicker product launches enabled by reusable, well-structured datasets
Client at a Glance
Service Type
Data collection & annotation for computer vision
Industry
Consumer electronics / smart security cameras
Engagement
Multi-year collaboration
Region
Global
Our client is a global consumer electronics brand producing smart indoor and outdoor security cameras.
Their portfolio includes multiple camera models with different optics, image sensors, and motion-detection sensors (PIR and radar). Some of these devices support on-device, edge processing, while others rely primarily on the cloud.
Challenge
The client needed to advance their computer vision capabilities across both edge and cloud environments.
Key Challenges:
Heterogeneous devices & processing modes
Different camera models had different optics and sensors. Some allowed on-device (edge) processing of raw video and sensor signals, while cloud-based models worked with compressed video (e.g., MP4). Algorithms had to perform reliably in both modes.
Insufficient public datasets
Available open-source datasets could not provide the necessary depth and diversity of scenes, nor did they align with the client’s strict KPIs for algorithm accuracy and inference time across edge and cloud.
Strict performance and resource constraints
New models needed to:
Meet accuracy KPIs
Be optimized for real-time processing
Respect the limited computational resources on the camera for on-device inference
To reach these goals, the client needed a large, custom, high-quality dataset and an efficient end-to-end data pipeline - from collection and consent to annotation, quality assurance, and reuse across product lines.
Solution
SQUAD designed and executed a multi-year data program covering legal alignment, data collection, annotation, platform development, and quality control.
Legal & compliance foundation
Aligned with the client’s Legal team on terms and conditions for launching a data collection program.
Prepared all necessary agreements to capture legal consent for collecting and processing video and sensor data.
Clearly defined the purpose of data usage: improving customer experience and product quality.
Large-scale data collection program
Launched a data collection program involving employees of the client and its vendors.
Created a specialized software solution for capturing and securely transmitting consented raw video and sensor data from deployed devices.
Added a cloud-based workflow to manage incoming consented recordings.
Custom cloud video annotation platform
We built a high-performance, cloud-hosted video annotation platform that brought several critical workflows into a single system:
Data collection & ingestion
Data annotation (video & sensor streams)
Task management
Dataset management & versioning
This enabled the client to scale annotation efforts while retaining full visibility into dataset composition and quality.
Operational processes & quality control
To ensure consistent, production-grade datasets, SQUAD:
Developed standard operating procedures (SOPs) for both data collection and annotation teams.
Trained a data quality team to run dataset quality checks and provide ongoing guidance to collectors and annotators.
Applied machine-assisted annotation to boost productivity while keeping humans in the loop for accuracy and edge cases.
Technologies & frameworks
The engagement leveraged:
High-accuracy object detection algorithms for pre-annotating videos and accelerating annotation.
SQUAD’s data annotation guides and special instructions for dataset quality checks.
SQUAD testing space with:
A dedicated motion-detection testing area
Multiple rented houses to capture diverse scenes, landscapes, lighting conditions, and environmental backgrounds
Tech stack: C++, Python, AWS, and custom tooling for:
Device firmware for data collection
The video annotation platform and supporting services
Results & Impact
Over several years, SQUAD and the client built a large, reusable dataset and transformed how computer vision was developed
and deployed across the product line.
technical outcomes
300+ TB of data
Video and sensor data collected for training and validating both edge and cloud algorithms.
Optimized resource usage
Edge models were carefully optimized to preserve limited compute resources on cameras, enabling additional features to run on the same hardware.
business outcomes
Faster time-to-market
By reusing well-structured datasets across similar product categories, the client could speed up product launches and feature updates.
Reduced Cloud costs
Moved most computer vision algorithms from the cloud to the camera, eliminating redundant cloud costs and enabling on-device inference.
customer outcomes
User experience
Increased satisfaction with features powered by computer vision (e.g., motion detection, alerts).
Response time
Reduced notification latency for camera-triggered alerts, thanks to on-device processing.
Contact us
by filling out
the form
to get started.
Get In Touch
Other Cases
Development and Optimization of Edge Computer Vision Algorithms
20+ edge CV projects delivered
Real-time multi-class motion detection
Improved motion detection rating
Fisheye Distortion Correction for Wide Angle Security Cameras
Consistent rectified video feed
Reduced geometric distortion
