Dec 12 – On account of computer vision, Amazon’s cameras can recognize products easily – without the use of barcodes. Amazon says that the system will eventually support robots.
According to Nontas Antonakos, an applied science manager in Amazon’s computer vision group, enabling robotic arms to pick up items and process them without requiring them to identify and scan a barcode is mission-critical. “It will help us get packages to customers more quickly and accurately.”
Amazon does away with the barcode
Due to the fact that robotic arms, used in Amazon warehouses, are not facile with barcodes, which can be hard to find and affixed to oddly shaped products in unexpected ways, Amazon intends to do away with the simple barcode.
Instead, a specially developed camera system can monitor items that move along conveyor belts, matching items with computer-catalogued images. In the future, Amazon’s AI experts and roboticists intend to develop technology that allows robots to identify items while lifting them up and manipulating them.
The system is known as multi-modal identification. While it isn’t going to fully replace barcodes in the near-term, it is in use across several Amazon warehouse facilities, according to the company. The process is reducing the amount of time required to process consumer requests.
New applications; computer vision
Experts believe that the technology could be shared across the entire Amazon business – meaning that you might see a version of it at Whole Foods, or within other Amazon-owned retail locations.
Computer vision powers smart devices
Previously, Amazon has build computer vision into its other merchandise. For instance, a user can ask an Echo Show smart display “Alexa, what am I holding?”. The tech is programmed to recognize home goods. The aforementioned feature, called Show and Tell, was designed in order to support the visually impaired.
Smart phone manufacturers and social media firms have also added AI features in cameras and photo apps, enabling the devices or programs to automatically categorize and sort photos.
Executing on computer vision
In order to execute on this project, Amazon’s AI experts needed to develop a repository of images of products. The images, along with data pertaining to products’ dimensions, fed the earliest versions of the algorithm, and the cameras consistently capture fresh images of products in order to train the model.
According to Amazon, the algorithm’s accuracy rate waffled between 75% and 80%, initially. Amazon saw this as a positive beginning. Now, the accuracy is at 99%. A glitch was identified when the system failed to catch color differences.
In short, Amazon’s artificial intelligence team states that fine-tuning the multi-modal identification system so as to identify products handled by humans will be problematic – Hence why the company is shifting towards robotic handling.