2017
OpenCV
YOLO
Deep Learning
Caffe
Torch
TensorFlow
Cnn
Openblas
Flare is smart home security device, which is using powerful machine learning to recognise owners from foe and take decisions by itself in case of danger. Flare is using an ensemble of algorithms to take reliable decisions.
Technologies
Caffe, TensorFlow, Torch, Yolo1, Yolo2, OpenCv, OpenBlas, CNN, Deep neural networks, GMM
Datasets
VOC, COCO, Celebrity 100, Faces in the wild, Multipie dataset, custom made ones, etc.
Components
1. Face detection
1. Haar Face/NPD/Yolo2 face detection
2. Motion filtering (keep an up to date background using Background Subtraction, and find regions of
interest (ROI) and motion likelihood)
3. Warm region estimation for the detector (most likely region that will contain an object at a given
time)
2. Face Recognition
1. Face feature extractor trained with Siamese Net with single channel grey scale images, (recently we enhanced our dataset with coloured images as well)
2. Face recogniser based on GMM (Multiple component GMMs for improved results)
3. Object detection
1. YOLO2 for pet, person, and face detection (Our object detection algorithm picks a model from several models of varying complexity (i.e. ranging from 0.29 MB to 5.6 MB) and analyse the ROIs and if necessary the full frame)
2. Optical flow estimations
3. Real time adjustments to object detection class thresholds (possible to immediately change the
person and face detection thresholds, based on adjustable alertness levels)
4. Speaker/Speech Recognition
5. Sound Recognition (speech, glass breaking, steps, fire alarm, dog barking, door knocking, etc.)
1. CNN with Mel-frequency cepstral coefficients (MFCCs) features
2016
Java
JPA
Azure
WebSocket
Redis
Twilio
Microsoft SQL Server
MQTT
Hibernate ORM
BuddyGuard's IoT Hub is the bridge between IoT devices, mobile apps and other API clients.
Technologies
Java 8, Spring Boot, Spring MVC, Spring Integration, REST Api, MSSql, MySql, Hibernate, Jpa, Redis, Distributed cache, MQTT, HikariCP, Azure Cloud, Azure Queues, Azure AppInsights, Azure notification hubs, Azure blob storage, Twillio integration, Junit.
Components
1. Bidirectional low-latency MQTT communication with mobile phones and IoT devices (infrastructure,
architecture of the communication, scalability and monitoring)
2. Synchronisation of IoT devices and mobile phones under various scenarios (no internet on device and
on mobile phones, differences in the speed of data transmission, unpredictability of device’s online
status, etc.)
3. Authorisation and roles system to cover various scenarios (multiple types of people/partners using the
system)
4. Authentication in the system in multiple manners (token-based login, fingerprint login, pin code login,
geolocation login)
5. Over the air update of IoT devices, coordinated by Cloud project
6. Billing system using several provides (PayPal, Stripe), for different services offered by company
7. Video peer to peer Livestream synchronisation using peer-to-peer technology for mobile phones and
firmware
8. Scalable storage mechanisms for raw data (image, video, audio, etc.)
9. Generic structures in project to allow fast model creation & testing. The project was really well tested,
4-5k+ tests
10. Special features : Geolocation auth, Security circle, etc.