In a LIVE AI/ML Masterclass, Prof. Anoop Namboodiri from IIIT Hyderabad, shared some interesting views and insights on Machine Learning in Computer Vision.

Computer vision is nothing but the eyes of artificial intelligence (AI). It is a technology that trains computers to analyze images, and this is done through the help of pattern/object recognition. Today it is being applied for various areas that straddle industries as diverse as healthcare, transport, BFSI, and retail. Some key applications cover:

 

  1. Securing high-value assets through real-time scanning and monitoring
  2. Identity verification and fingerprint analysis 
  3. Broadcasting, sports with virtual replay and image overlay
  4. Virtual and augmented reality
  5. Motion recovery and reverse engineering
  6. Security of banks, research labs, safety boxes, elevators, offices, or high-value archives
  7. Biometric analysis in security
  8. Quality inspection in manufacturing
  9. Automated assembly
  10. Mail sorting
  11. Robot navigation
  12. Monitoring industrial processes and detection of non-conformities
  13. Self-driving vehicles with LiDAR sensors
  14. AI-aided surveillance for cranial imaging
  15. Monitoring of acute neurologic events and detection of neurological illnesses
  16. Tracking and warning against obstacles
  17. Reading road-signs
  18. Guiding emergency vehicles like fire safety trucks and ambulances
  19. Content-based image retrieval
  20. Food and packaging processes 

Computer vision is getting more accurate and intelligent than human vision, thanks to various technologies like deep learning, facial recognition, and scanning.

Computers or humans? or both?
For years, computer vision powered by robust machine learning (ML) algorithms has fueled the curiosity of entrepreneurs, scientists, and engineers. From face recognition to processing the live action of a cricket match, it rivals human visual abilities in many areas. But, how does it seamlessly hand-shake with machine learning to solve real-world challenges? 

Automated vision is characterized by high reliability, high repeatability, more objective evaluation, lower cost, higher speed, and the ability to operate in hazardous environments. And why are we trying to emulate human vision – well, because it is the best genre we know.

Master class on “Learning to See: Machine Learning in Computer Vision” by Prof. Anoop Namboodiri, Associate Professor at IIIT Hyderabad  

Should computers process images like human beings? Not always, as experts opine. Human vision takes some short-cuts to figure the world around us. But the human vision also has its limitations. The brain can play tricks with our eyes and give optical illusions. Computer vision cannot be fooled. But human vision has a lot of learned information which is used for interpretation. This is both a gift and sometimes an impediment.

It is interesting to note how computer vision would find scope or constraints here. The urge to group, recognize and measure are three fundamental aspects of the human brain in vision. What is easy, obvious, and involuntary for a human eye is not simple for a computer. Mostly for a computer, all this is an array of numbers. The computer uses them to figure out shape, context and build up information in a bit-by-bit manner. The information is present at a lower level. The human high-level of information is involuntary because it comprises several years of learnt information.

Can we make computers deploy inference and humans do? That isn’t easy because we do not know how we infer anything. 

That’s why we have to look at creative solutions. Like taking help from Geometry for solving the ‘measure’ problem. Taking help from learning for the ‘segmentation’ problem or combining computation and geometry for other answers. Using 360-degree vision is also an advantage here.

A lot of work and innovations are making something apt for Indian roads and Indian constraints. All these are necessary and encouraging areas of progress. It is affecting the rise of the market on many fronts.

Laser-sharp and laser-fast
The global computer vision market stood at $10.6 billion in 2019, according to estimates from GrandView Research. Moreover, it is slated to grow at a CAGR of 7.6 percent from 2020 to 2027. Interestingly, the Asia Pacific segment made up for over 38.0 percent share of global revenue in 2019.

The speed and scale of this vision are staggering, but a lot of work needs to be done on inference and context so that costly mistakes, false positives, and bias are avoided.

As these eyes get sharper and faster, they will see more and show more than what we see through human eyes. The world would look different with these glasses on. Very different and powerful. To build relevant capabilities and explore such possibilities, IIIT Hyderabad and TalentSprint offer PG Certification Program for Mid-to-Senior Tech Professionals. This unique program comes with a curriculum specially designed for practitioners, lectures by distinguished IIIT Hyderabad faculty, industry sessions by eminent professionals, labs supported by mentors, group projects and hackathons, and more.