Company
vsys offers custom computer vision solutions — from prototype to production (since 2016)
Domain
- CAMERAS & SENSORS
- pinhole and fisheye cameras
- RGB-D and stereo imaging
- multi-camera systems
- LIDAR
- IMAGE PROCESSING
- image processing tasks
- optical flow
- background extraction
- image registration tasks
- panorama imaging
- 3D VISION
- camera calibration
- stereo vision
- 3D reconstruction
- SLAM, graph optimization
- bundle adjustment
- point cloud processing
- Gaussian Splatting
- DETECTION & TRACKING
- 2D/3D object detection/tracking
- DEEP LEARNING
- custom neural networks
- YOLO networks
- training data simulation (Blender)
- motion capture systems
- PERFORMANCE
- real time and offline solutions
- GPU (CUDA, OpenCL, GLSL shaders)
- performance optimization
- TOOLS & LANGUAGES
- C++, Python, numpy, PyTorch
- OpenCV, OpenGL
- visualization
- numerical optimization
- custom algorithm development
- PLATFORMS & DEPLOYMENT
- Linux, Windows
- embedded systems, IoT
Selected projects
- Computer vision system for pool
- A real-time VR controller pose tracking algorithm
- Dense 3D reconstruction using moving monocular camera
- 2D ellipse detector for bin picking
- Upgrading stereo vision of a food delivery robot
- Real-time panorama stitching for food delivery robot
- VR controller detection from image using HoG
- HDR conversion for RGB-D based 3D mesh
- Low-cost computer vision system for tennis
- Stereo vision for autonomous driving
- Content pipeline from LIDAR to Unreal Engine
- Camera based car counter for parking lot monitoring
- Human 3D reconstruction using RGB-D sensor
- 2D trash tracking for analysing robot picking
- Plane sweep method for multi-view depth map generation
- Interactive and scalable 3D point cloud visualizer
Upgrading stereo vision of a food delivery robot
GPU implementation of semi-global block matcher for improving stereo depth map quality.
Discover
Real-time panorama stitching for food delivery robot
Real-time panorama stitcher implemented using CUDA and OpenCL to Nvidia and Ryzen platforms.
Discover
VR controller detection from image using HoG
Custom HoG detector trained using random forest and implemented in C++ to detect VR controller from 2D image.
Discover
HDR conversion for RGB-D based 3D mesh
Upgrading 3D reconstructions into HDR space by modeling consumer camera exposure and gain mapping into unified color range.
Discover
Low-cost computer vision system for tennis
Ball and player tracking for tennis with automatic calibration and spin detection. Platform: Yocto Linux.
Discover
Stereo vision for autonomous driving
Stereo camera is setup on car rooftop for real-time depth perception. Depth image based object detection is also prototyped.
Discover
Content pipeline from LIDAR to Unreal engine
Preprocessing pipeline to convert colored LIDAR point clouds into compressed and textured mesh format.
Discover
Camera based car counter for parking lot monitoring
Real-time camera based car counter using optical flow on gates.
Discover
Human 3D reconstruction using RGB-D sensor
Prototyping human 3D reconstruction using RGB-D sensor.
Discover
2D trash tracking for analysing robot picking
Tracking trash feature points to solve the problem: What was picked?
Discover
Plane sweep method for multi-view depth map generation
Generating a depth map using multiple calibrated views with high-reso images
Discover
Interactive and scalable 3D point cloud visualizer
OpenGL based point cloud renderer using octree structure for traversing large point clouds
DiscoverContact
Have a project in mind? Describe it below and I'll get back to you within one business day.