Embedded360 develops state-of-the-art on-device artificial intelligence (AI) and vision solutions that enable businesses to make real-time decisions and deliver more efficient experiences to customers. From building extremely low footprint and low power models on constrained end-points to bringing full-fledged, high-accuracy algorithms on edge devices, we bring the right solutions for our customer’s applications. In the last decade, we have delivered AI and vision solutions, based on traditional machine learning as well as contemporary deep learning algorithms, for applications spanning security & surveillance, robotics, AR/VR, drones, automotive ADAS & AD, and industrial domains.

Data management

• Data preparation for both training and validation
• Data with variation to cover all possibilities
• Data annotation and enhancement

Model design and training

• Complexity aware custom architectures including traditional ML and FF-CNNs, RNNs, LSTMs, CRFs, HMMs
• Selecting the right frameworks
• Model tuning and compression

Embedded optimization

• Real time implementation of networks on embedded processors
• Embedded friendly network pruning, float-fixed point conversions and hyper parameter optimization

Benchmarking & validation

• Measurement of inference metrics such as performance, power, memory etc.
• Utilize frameworks, tools, automation scripts for effective validation

High accuracy

We build high accuracy models, under the constraints of processing power and memory availability

Fast performance

We leverage our deep understanding of underlying processor architectures to squeeze out the fastest performance in milli seconds

Small footprint

We build models with tiny memory footprint so that more applications can run on even the most low-end hardware

Re-usable

We build modular algorithms that can be re-used across platforms and architectures for your product revisions

Superior implementation

We build application specific models and use application specific data to make sure that you get an optimum solution

Quick deployment

Leveraging our existing building blocks and understanding of AI architectures, we cut down your development time by at least 30%

Facial recognition & analytics

A state-of-the-art computer vision algorithm that can detect and recognize faces with wide pose and illumination variations. It also enables demography detection and emotion detection. Available in multiple complexity variants that can run on a range of underlying platforms

Automotive ADAS & AD

A set of algorithms for developing camera-based ADAS and AD solutions. Currently includes models for traffic sign recognition, traffic light recognition, vulnerable road-user detection, road-lane markings and semantic segmentation. Available on major automotive grade platforms

Radar interior sensing

Machine learning based algorithm suite for in-cabin sensing applications such as child/adult/life presence detection, seat belt reminder, driver out of position detection and gesture recognition. Works on low cost sensors and embedded processors.

Driver Monitoring

Complete SDK for driver monitoring (identification, drowsiness, in- attention) and in-cabin sensing (occupancy detection, classification and more). Also provides APIs for eye tracking, facial analysis and activity classification. Available in multiple variants