How is AI enabling “Object identification and recognition”?

Object identification and recognition is an area within the field of artificial intelligence (AI) that focuses on robots recognizing different objects. As AI machines become more and more part of our everyday lives, machine learning is making improvements on image tagging and object identification skills. Google and Microsoft are examples of just two tech giants investing in object identification tech.

 Computer Vision Basics

Computer vision is the the science of computer systems recognizing and analyzing different images and scenes. A key component of computer vision is object detection. Object detection is used to perform a number of AI tasks such as facial recognition, vehicle detection, security scanning, and self-driving. A number of algorithms depend on object detection to work successfully in AI implementations. The most well documented algorithms tied to object detection processes include R-CNN, Fast RCNN and Faster RCNN. The CNN in the algorithm stands for convolutional neural networks. CNN is designed to focus on pixels within images located next to each other. Each image is passed through the network as an input and then sent back as an output with each object classified.

Recently, a number of advancements in deep learning have paved the way for solving object detection problems. MIT has funded a number of object detection projects including the development of neural networks based on how the human brain functions. Deep neural networks have shown a high-level of performance for successfully completing object classification tasks. MIT also created a deep learning system that allowed for object identification to occur in real-time through speech recognition.

Object Detection Challenges

A challenge within AI systems is to employ accurate object detection when in contact with fast-moving objects like vehicles. In July of 2018, researchers from the University of California created a 3-D printed neural network with the capacity to use light photons to rapidly process images and recognize certain patterns to enable object identification. The AI network had over 91 percent accuracy for object detection with the research team continuously working to improve performance.

Another challenge for AI is that studies are demonstrating fallacies within current systems. For instance, Auburn University released a November 2018 research paper that Inception, Google’s image-recognition system could be tricked by object rotations. As an example, a school bus was placed on its side by the research team with Inception incorrectly identifying it as a snowplow. The point of the study was to show Inception could be easily confused with out of the norm placement of objects.

Improvements in object detection are critical to the future of AI. If robots can understand their surroundings better, they are able to perform complex tasks.