Computer vision is opening up possibilities of adding differentiating capabilities in high-tech products by enabling these products to see and understand the world around them better. However, computer vision is computationally intensive for typical embedded systems that the high-tech products run on – it consumes prohibitive amount of resources including CPU and memory. Add to that the challenge of achieving accuracy requirements and it becomes a huge adoption barrier. PathPartner, with its extensive expertise in imaging/camera sub-systems, deep learning methodologies and real-time embedded systems, enables customers to harness the power of embedded vision technologies in their next-gen products. With our full stack services from selecting the right embedded platform to developing CNN algorithms and from camera integration to optimizing the systems for embedded systems, PathPartner is your ideal partner in your embedded vision integration journey.

Looking to embed computer vision in your products ? Explore our end-to-end offerings

Image capture
  • Camera hardware and software for image/video capture
  • Image quality tuning and enhancements, Camera driver development
Data Preparation
  • Dataset collection from multiple sources
  • Data annotation & verification
  • Problem specific data augmentation
Model design
  • Complexity aware custom architecture design including FF-CNNs, RNNs, LSTMs, CRFs, and HMMs for use-case
  • Framework selection
  • Model training
Model optimization
  • Algorithm porting and optimization on embedded platforms including deep compression
  • GPU acceleration
System integration
  • System software development including connectivity, display integration
  • System optimizations for constrained resources
Application development
  • Application development for automotive, retail, robotics, drones, surveillance, security, industrial use-cases

Building blocks for faster development

Use PathPartner’s proven, reliable, and ready-to-integrate hardware and software components in your next product and accelerate your time to market

Plug-n-play camera modules
Face detection and recognition
Eye gaze tracking
Object detection and classification
Object motion and tracking

Platforms. Frameworks. Tools. Check.

Platforms

Embedded platforms encompassing ARM, DSP, Vision accelerators, FPGAs, Custom SoCs and camera sensors

Frameworks

Caffe/Caffe2, Tensorflow
OpenCV, OpenVX, OpenCL

Partnerships and proceses

Computation backend – CUDA, OpenBLAS Data visualization : Pandas, Tensorboard, Matplotlib
Stats workbench : Python

Target Applications

Automotive

Driver Monitoring
Traffic Sign Recognition
Obstacle detection

Robotics

Collision avoidance
3D reconstruction

Retail

Audience classification and measurement

Gaming, AR, VR

Eye-gaze tracking
Smart glasses
VR headsets

Others

Surveillance
Video conferencing

Our expertise in action: PathPartner’s Driver Monitoring solution

PathPartner’s driver monitoring solution, uses advanced deep learning methodologies to assess drivers’ alertness under diverse conditions. It uses a hybrid combination of advanced facial analysis algorithms across 28 facial landmarks to perform real time analysis of driver’s eyes, mouth, and head movements. With a thorough study on each of these problems complexity, deep understanding of deep learning models’ representation abilities, we developed algorithms that can run real time on embedded systems (Ex: our DMS can run real time on Qualcomm's SD820 and NVidia TX2)

Know More