Machine Learning and Computer Vision : A Dynamic Duo on a Mission

One of the most intriguing concepts for Computer Scientists was to enable computers to "see" and comprehend their environment. The fiction of yesterday has become the fact of today. All these have been possible because of the advent of Artificial Intelligence with whose communion Computer Vision has taken a huge leap towards integration in our daily lives .

In this article , we will review the concepts of how the duo of Machine Learning and Computer Vision is on a mission to transform various technologies and bring our perception to precision . Automation is being brought to a number of sectors that people could not have even dreamed of a decade ago , thanks in large part to Computer Vision. Businesses in a wide range of sectors are being significantly impacted by computer vision, including those in retail, security, healthcare, construction, automotive, manufacturing, logistics, and agriculture.

How Machine Learning is revolutionizing Computer Vision ?

Deep Learning is a subset of Machine Learning which is itself a subset of Artificial Intelligence . Deep Learning involves various algorithms of Neural Networks whose basic building blocks are Perceptrons (artificial neurons ). The development of Convolutional Neural Networks(CNNs) represented the crucial turning point . Initially an architecture called LeNet-5 was developed which was later improvised and refined to make AlexNet . It brought great heights into the accuracy of object classification and detection .

CNNs helps in pattern recognition which enhances object detection . CNNs has mainly 3 layers:-

1. Convolutional Layer

2. Pooling Layer

3. Fully Connected Layer

Since convolution is a linear operation and images are far from linear, non-linearity layers are often placed directly after the convolutional layer to introduce non-linearity to the activation map. Some of them are Sigmoid,Tanh and ReLU.

Where can we apply Computer Vision Technology?

• Facial Recognition

  Facial Recognition is now being used in many places such as phones,places where biometric authentication is required. Here generally 2 convolutional layers along with 2 pooling layers are used just like the LeNet-5 model.

• Augmented Reality(AR)

  It is the field where physical environment intermingles with the digital or the virtual world. AR uses SLAM(simultaneous localization and mapping), a computer vision algorithm that compares visual features between camera frames in order to map and track the environment . AR needs specially designed Machine Learning models which involves lesser computations as it needs to run on the Edge . Thus optimised neural networks must be used to run the model in real time in a phone.

• Health

  Computer Vision has most of it’s contributions in the field of healthcare . It incorporates complex CNN architectures to study a huge group of data pattern and later classify and detect the disease and also analyse it’s severity.

  Cancer Detection : Machine Learning in medical industries helps the doctors to identify skin and breast cancer majorly using slight changes between cancerous and non cancerous through image recognition. It also analyses the magnetic resonance imaging(MRI) scan data using deep neural networks for high accuracy.

  COVID-19 diagnosis : Multiple deep learning computer vision models exist for x-ray based COVID-19 diagnosis. The most popular one for detecting COVID-19 cases with digital chest x-ray radiography (CXR) images is named COVID-Net and was developed by Darwin AI, Canada.

  Disease Progression Score : Deep Learning in Computer Vision has come to such an accuracy that it can properly diagnose based on the disease , it’s severity and the amount of extra care that has to be taken.

  Movement Analysis : There are various neurological and musculoskeletal problems such as osteoarthritis , balance and gait problem which can be detected with the help of Deep Learning without the interference of the doctor . People can use these for Pose Estimation for analysing themselves and maintain a healthy life . But this needs Edge Computing as it must happen in real time. In this way it can be incorporated even in the sport practices .

• Agriculture

  Agriculture is one of the most arduous works , thus automation in that field is highly appreciated

  Plant disease detection : Saving plants from disease is very important for proper yield . Thus with the advent of CNN we can detect plant disease even by avoiding the labour-intensive feature engineering and threshold-based image segmentation processes.

  UAV Farmland Monitoring : Real-time farmland information and understanding that information plays a major role in agriculture . This system is brought to face by the drones having high resolution camera and an array of sensors to detect each and every details of the agricultural land. Irrigation can also be managed by this system.

  But for all these to happen again Edge Computing has to be done without sending anything to the cloud for giving real time detections and analysis.

• Transportation

  Traffic Analysis : Traffic flow analysis is a part of the smart city plan where with the rise of AI and computer vision , video analytics can now be applied to ubiquitous traffic cameras . The traffic flow can be observed using computer vision means and measure some of the variables required by traffic engineers.

  Self driving Vehicles : The greatest achievement of Computer Vision . The car learns to drive it by avoiding obstacles using the obstacle detection code and also it is most likely not to repeat again . This is because in those Reinforcement Learning is also incorporated.

• Image Captioning

  It is fairly new and is undergoing development everyday . It aims to generate descriptive legends for image but there are challenges to understand the proper language both syntactically and semantically , here Computer Vision intersects with Natural Language Processing . Till now it is found that GANs(Generative Adversarial Networks) based models give the highest accuracy.

• AI Image Generator

  The best examples of Text to Image are Stable Diffusion, Dalle-2 and DeepAI where they use GANs , firstly the models get trained over a large set of images then unique images are generated by the AI, mainly by using StarGAN where in each and every step some noise are added to produce a completely unique image.

All these functions are used together in Robotics industry to give the robot the ability to see and understand the world and act accordingly.

What’s next?

Despite the recent impressive progress , there is a lot more to achieve in this field . Computer Vision is yet to be explored extensively with technologies like VR(Virtual Reality) and also in several other domains of works. Visual Data being the best ever possible data , needs much more exploration as it is getting created each and every second, which people think as a curse of our generation , are actually used for our benefit .