Home camera imaging computer vision

Computer Vision Experience

Over 10 years of researches in the Computer Vision domain with experience in both PC-based CV algorithms development and Embedded platforms porting.

  • FPGA, DSP and GPU CV methods acceleration skill set
  • Production-grade Audience Measurement video-analytics that includes: bidirectional people counting, face detection & tracking, facial features classification (gender, age group, pose)
  • Integration of CV algorithms of Ambarella’s CV partners into camera solutions designed by Rhonda Software
  • CV-powered Factory Image Quality calibration and Optical systems inspection procedures:
    • Automated IQ calibration results validation
    • Image sensor dirt contamination detection
    • Sensor to lens alignment check
    • Optical resolution calculation powered by automatic chart detection
    • etc.

Our video demos

People counting with top-mounted camera
This people counting solution for bidirectional traffic uses top-mounted camera. It's based on more common technology of line crossing detection, supports multiple lines and can be used to count not only humans but any moving objects like vehicles when they cross user-defined line placed in any part of a frame.

Tools recognition
This pattern recognition solution is based on user-defined learning samples. Sample application is pre-trained to recognizes 5 types of tools (hammer, plies, pipe dog, screwdriver and wrench sockets) by their specific silhouettes and, what is more, measures inner diameter of sockets. Several tools in any orientation can be recognized at the same time.

USD banknotes recognition
This pattern recognition solution works with all designs of USD banknotes with all denominations from $1 to $100; recognized banknotes in any orientation as well as partially overlapped and folded banknotes. Sample application counts money grouping banknotes by their denominations.

Barcode recognition (mobile platform)
This solution is developed especially for resource-limited mobile platforms, it detects and recognizes one-dimensional barcodes of any sizes and in any orientation. Sample application works on Windows Mobile with 520 MHz and allows adjusting camera resolution.

Object Recognition (Nike logo)
This highly tailored solution works with only one pattern; it recognizes Nike logos of any sizes and colors. It's possible to recognize any number of Nike logos in any orientations at the same time, as well as to detect and recognize Nike logo under specific conditions – partially blocked logos and logos placed of crumpled surfaces like paper or fabric.

Audience Measurement
This audience measurement solution measures demographics, impression, opportunity to see, vehicle traffic, dwell time and other standard audience measurement metrics. Sample application marks male faces with blue rectangles and female faces with red rectangles distinguishing those visitors who are actually looking at camera (bright color of rectangular for attention) and visitors not looking at camera but with visible faces (pale color for presence).

Barcode recognition
This barcode recognition solution works with one-dimensional barcodes of any size and orientation including partially blocked barcodes.

Tracking overlapping objects
This solution illustrates color-histogram-based object tracker in action. Sample application tracks people as moving blobs (“clouds” of moving pixels) and uniquely identifies them. When two or more blobs are overlapped, sample application merges them into one combined blob and marks it with IDs of all objects included. When one of objects separates from this blob sample application recognizes which one is out and re-arranges ID appropriately. This approach works pretty well in case of characteristic histograms.

Object Detection (barcodes)
This is the demo of the jerry-built algorithm that finds barcode plates using Hough transform. Actually the video is mostly speaking for itself.

Taking snapshots with a moving PTZ camera
This solution uses two cameras – moving PTZ camera and static video camera. Sample application tracks moving people (using color-histogram-based tracker) on a video taken with static camera and targets PTZ camera to one of people for snap shooting. Since PTZ camera positioning takes some time predicting algorithm is used to forecast future position of moving person; it helps targeting PTZ cam more accurately.

Object Recognition (playing card & $10 bill)
This object recognition solution is based on technique of extracting and matching characteristic points of objects with complicated texture. Sample application allows recognizing two types of objects – playing card (queen of hearts) and $10 bill. It's possible to recognize partially overlapped object under various view angles and orientation including partially overlapped objects.



Do not hesitate to contact us for any question regarding our services. We will be glad to respond in detail.