For a long time “PAWLIN Technologies” company develops algorithms and software for automatic selection of objects in images using a number of computer vision and pattern recognition methods.
Goal
Automate the process of manual selection of specified classes in the image to provide the customers products with a new functionality or to increase the productivity of operators involved in manual image processing.
Input data
The customer provides a representative sample of images containing and not containing the specified objects, or formulates different description of specified objects, for example, providing single standards. Optionally, the customer provides a priori information about the desired objects — the degree of variability of their shape, subclasses, possible lighting conditions, camera angles, etc. It is possible to manually create a training sample by company specialists using open sources or customer’s materials. In some cases, it is possible to generate a training sample using an imitation model of objects themselves or to increase their number by artificial changes in their shape.
Output
As a result, the algorithm and its software implementation, that ensures the task of detecting a given class of objects in images, are developed. In order to achieve certain processing speed requirements, can be applied acceleration of computations on graphic or multi-core processors. A software implementation for single-board computers (embedded) is possible.
Examples of developed detectors:
-
- search for reference points of a specialized calibration board, ensuring unique determination of its orientation;
-
- peoples faces detector;
-
- search for logos (trade marks) in video stream frames and photo images;
-
- search for similar frames (for pirated video);
-
- search for strange inclusions in frozen food products based on analysis of X-ray images (flaw detection);
-
- human head and shoulders detector;
-
- car detector (in aerial photos);
-
- road detector (in aerial photos);
-
- rectangular objects detector (in aerial photos);
-
- ships detector (in aerial photos);
-
- animal (moose) detector (in aerial photos);
-
- flags and cards detector (in robotics);
-
- digital signature detector;
-
- cloud detector;
- forgotten objects detector.
For specialists
Caffe; CNN; MLP; MSER; SIFT; SOM; Connected Components; GPU; OpenCL; OpenMP; OpenCV