In essence, computer vision is a subset of artificial intelligence that enables computers to see and make sense of the world. It is the AI application that allows a computer to learn to analyze information from photos to video to thermal and infrared data, amongst other sources, and then make decisions or come to a clear understanding of the environment or situation based on that information.

Computer vision is an incredibly sophisticated tool. The best way to appreciate this complexity is to consider how human sight works. Human eyes have some six to seven million cone cells, each of which contains one of three color-sensitive proteins, called opsins. When light photons make contact with these opsins a reaction occurs that generates electrical signals. This message is transmitted to the brain where it’s interpreted.

Building an application capable of replicating that process is phenomenally intricate. Such a system would have to be capable of processing and interpreting visual information so that it can be used for pattern and object recognition, and for adapting 2D images from our 3D world into 3D information. Yet in just a decade, the accuracy of object identification rates has increased from 50% to an incredible 99%.3 This makes computer vision more accurate than humans at reacting rapidly to visual data.

Computer vision is based on deep learning, a form of machine learning that trains computers to perform human-like tasks. Currently, these broadly include understanding natural language, identifying and classifying objects in images or making predictions, but other uses are evolving. Rather than running data through a series of predefined equations, deep learning establishes the structure of the model through some basic parameters concerning the data, then trains the computer to learn on its own by recognizing patterns using many layers of processing.

To emulate human sight, machines need to acquire, process, analyze and understand images, which is made possible due to the iterative learning process of deep neural networks.

The process begins with a curated set of images or video data, known as training data. This is used to help the machines learn certain things about a certain topic; think damage to vehicles for the insurance industry or prohibited luggage items for the airline industry.

Any training data will require within it sample images of luggage containing prohibited items, such as aerosols, weapons, and liquid containers, as well as those containing permitted items. Each image will be tagged with metadata indicating the correct answer – in this case, permitted or prohibited.

A neural network will process the visual data, using pattern recognition to identify the many different components of an image. Its outputs, or ‘answers’ as to whether an item is allowed or not, are fed back into the system allowing it to learn and improve in accuracy. So, instead of a human attributing certain characteristics to items, the machine learns from the images it receives.

Achieving Higher Accuracy & Real-Time Performance Using DeepStream 

Managed IVA Apps & Services

For a real world IVA app/ service deployment, remote management and control of applications is critical. DeepStream SDK can run in any cloud and at the edge which makes it a powerful SDK to handle IoT requirements such as effective bi-directional messaging between edge and the cloud, security, smart recording and Over-the-Air AI model update.

• With bi-directional messaging between edge and cloud , you can add greater control for use-cases such as remote triggers for event recording, change operating parameters and app configurations or request system logs.
• The smart record feature in DeepStream app allows you to save valuable disk space on the edge with selective recording that enables faster searchability . You can use cloud-to-edge messaging to quickly trigger recording from the cloud.
• Seamless Over-the-Air (OTA) update for the entire app or individual AI models from any cloud registry to continuously improve accuracy with zero downtime.
• For secure IoT device communication , DeepStream provides two-way TLS authentication based on SSL certificates and encrypted communication based on public key authentication.

DeepStream offers an IoT integration interface with Kafka, MQTT and AMQP and turnkey integration with AWS IoT and Microsoft Azure IoT.

You can build high performance DeepStream cloud native applications with NVIDIA NGC containers. By using DeepStream, you can deploy at scale and manage containerized apps with Kubernetes and Helm Charts.

Amazon Rekognition

Automate your image and video analysis with machine learning.

Amazon Rekognition makes it easy to add image and video analysis to your applications using proven, highly scalable, deep learning technology that requires no machine learning expertise to use. With Amazon Rekognition, you can identify objects, people, text, scenes, and activities in images and videos, as well as detect any inappropriate content. Amazon Rekognition also provides highly accurate facial analysis and facial search capabilities that you can use to detect, analyze, and compare faces for a wide variety of user verification, people counting, and public safety use cases.

With Amazon Rekognition Custom Labels, you can identify the objects and scenes in images that are specific to your business needs. For example, you can build a model to classify specific machine parts on your assembly line or to detect unhealthy plants. Amazon Rekognition Custom Labels takes care of the heavy lifting of model development for you, so no machine learning experience is required. You simply need to supply images of objects or scenes you want to identify, and the service handles the rest.

FEATURES

Custom labels

Text detection

Face search and verification

Pathing