Capabilities: What objects we detect
Traditional video analytics and machine vision solutions work well with standard objects and clear backgrounds. For example, if we deal with a machinery part at a clear production line or if there’s a separately standing person – standard detectors allow reliably extracting these objects, counting them and doing other operation. But if this person just sits down, standard detector can’t classify such an object any more. Same situation if there’re overlapping objects moving around. Here BitRefine Heads comes to scene.
BitRefine Heads utilizes power of neural networks that comprehend visual information not as a rule-based detector, but similar to a living brain. For example, standard optical character recognition (OCR) may not recognize all the digits on a label, if it is partly deformed. At the same time, human inspector may still be able to read all the digits. And if human can read them, then we can be sure that ANN-based algorithm at BitRefine Heads is also capable to read them.
BitRefine Heads excels at exploring complex scenes.
We work at the domains where traditional detectors don’t work. Here're typical examples:
General objects
Our platform allows detecting general everyday objects captured from any angle, located in any environment, overlapping each other.
Low quality images
Perfect source images allow using simple detectors. But the worse quality of the image, the more complex algorithms required to classify objects. BitRefine Heads works is able to effectively process sources of low quality.
Aerial images
Cars, houses, roads, trees, lakes – these objects can be recognized by neural networks. The aim of such system is usually to collect statistics from large areas.
Biology
Microscopic imagery, MRI, ultrasonography, endoscopy – all these diagnostic methods provide image that are successfully analyzed by artificial neural networks.
Inspection of materials
Neural networks are perfect for recognizing and classifying defects, like scratches, wrinkles, corrosion, cracks, welding defects, etc. Typical applications are inspections of pipelines, solar panels, at textile production, metalworking.
Different spectrum images
BitRefine Heads works equally well with images captured with regular cameras and images captured in different spectrum, such as thermal images or X-Ray images. Separate neural modules, trained on different image sets are loaded it into the BitRefine platform’s “head”.
Animal or plant species
Neural network can distinguish birch from oak as good as human can. It also distinguishes perfect fruit from all others or detect defects on a fruit.
Sports
Proven ability of deep learning to detect person despite of angles of view or body position allows tracking sportsman and collect statistics during the game.
Segmentation
Special group of neural network architectures are capable to return coordinates or even shape of detected object in addition to its class.
Featured Articles
Object recognition software for quality control and worker safety
Traditional Analytics
Traditional analytics relies on formal object specification that includes measurable values: dimensions in pixels, aspect ratios, angles, colors and histograms. This means that the environment needs to be clear, parameters of detected objects need to be well documented and lie within defined thresholds.
If you’re dealing with well described parts, passing in front of the camera always at the same orientation and same light condition you don’t need to give up traditional computer vision approaches in favor of modern deep learning. Traditional algorithms work better and much faster than anything else in this case.
BitRefine Heads
If environment is too dynamic, if object shapes are too complex, if objects are randomly orientated – solution based on artificial neural network is the right choice. Artificial neural networks along are way more powerful than standard detectors and offer recognition solutions in the areas unreachable to traditional analytics.
The advantage of BitRefine Heads platform is that it fuses traditional methods with advanced capabilities of artificial neural networks. Platform offers numerous modules that can be inserted into the video processing pipeline during setup. Some of the modules are based on neural network algorithms, others belong to traditional filters or detectors. User is free to configure video processing pipeline his own way: to include and exclude modules, to combine and reposition them – to configure BitRefine Heads according to his specific needs.
Want to learn more?
Have questions? Contact us: