An East End technology company has spun out a startup that makes tools to help computers “see” and identify objects, with applications that could include guiding self-driving cars and cashierless convenience store checkout.

Innotescus LLC, which was born as part of Point Breeze-based ChemImage Corp in 2019, uses video annotation to “teach” computers to recognise objects such as traffic signals, backpacks and pedestrians. The company, named for the Latin phrase that means “we make known,” recently began marketing its first product.

The automated and semi-automated tool trains computers to identify an image that may appear in a number of similar forms in a video, said CEO Tim Tannert. A backpack might look different depending on the angle, for example.

“We think there’s a cool story here because Pittsburgh is starting to become a hub for technology and robotics and computer vision,” he said.

The privately held company has 15 employees. Competitors include London-based V7 Labs and Labelbox Inc of San Francisco, but Mr. Tannert said Innotescus may be the first of its kind in Western Pennsylvania.

Competing video annotation gear analyses each frame in a video so it can identify a given image. At 60 frames a second, the job is labour intensive at best and overwhelming at worst.

Innotescus leaders believe their advantage in the market is much faster annotation through the use of artificial intelligence.

Using conventional annotation software, a month is needed to mark up 2,000 frames of video. Innotescus’ software allows analysis of 40,000 frames with the same labour and time period with greatly improved accuracy, Mr Tannert said.

There’s growing demand for these kinds of services in the market. The marriage of video and machine learning is creating an array of new business applications, and that has also expanded the need for identifying and annotating objects in videos. Seattle-based e-retailer Amazon was an early pioneer in the field with the opening of cashierless convenience stories in 2018.

Amazon Go stores allow shoppers to buy fresh produce, fish, meat and ready to eat meals without checkout lines. The company’s ‘just walk out’ technology uses a swarm of cameras to detect merchandise that shoppers pick up, with purchase prices simply added to the customer’s Amazon account.

The stores stock 5,000 unique items, which must be identified by video taken by store cameras. Innotescus software streamlines the task of “teaching” machines to identify objects by using artificial intelligence algorithms.

“The world is being powered by machine learning and algorithms,” the company’s co-founder and CTO Chris Anderson said. “What really powers these tools is data. Data visualisation — we bring the data to life.”

Lawrenceville-based RE2 Robotics, a maker of robots used in construction and other applications, is among Innotescus’ customers. The software is used to help “perceive the world with computer vision,” said Amanda Sgroi, director of computer vision and autonomy.

“Having a tool like this in Pittsburgh supports the robotic community as a whole,” Ms. Sgroi said. “Roboticists need this type of data to interact with the world.” – Pittsburgh Post-Gazette/Tribune News Service

Source link

Leave a Reply

Your email address will not be published.