With pedestrian fatalities on the rise in recent years, San Jose, Calif., is working with Sony Semiconductor, TinyML and others to sponsor a developer hackathon intended to produce novel ways to use vision technology and machine learning to spot pedestrians and reduce accidents.
One idea is to use the latest small cameras with edge processing ability to activate flashing of nearby streetlights and signals when a pedestrian is jaywalking or otherwise entering improperly in the roadway.
The safety approaches could be as diverse as the various developers come up with, according to hackathon organizers.
Mayor Matt Mahan kicked off the hackathon in July in a tweet, with entries due by mid-September. Cash prizes up to $3,000 will be awarded in October, with city engineers serving as judges. “Please come help us make San Jose streets safer,” the mayor said. Pedestrian fatalities have doubled since 2012, a phenomenon seen in many cities who have joined a global Vision Zero initiative to prevent accidents.
The broad challenge of the hackathon, which has already attracted 20 developer teams, is to build a TinyML based solution to detect pedestrians and others. It notes: “People often do not walk in the crosswalk.”
Devkits are being offered by Sony, Infineon and Brainchip. A free Edge Impulse license and support are provided.
Sony’s devkit works off Sony’s new Aitrios edge AI sensing platform, which is being trialed by retailers and other companies.It relies on a chip and an image processor with AI processing on the backside to reduce power and networking needs, according to Mark Hanson, vice president of technology and business innovation for Sony Semiconductors Solutions America.
Images of actual people or their faces are not stored to protect privacy and AI is used to make a quantitative interpretation of human shapes, he said. “We jettison the images of the person. Aitrios moves the image from a pixel chip to a logic chip on the backside,” he said.
In addition to potential use for traffic safety, the smaller Aitrios cameras can be put to use in reading barcodes in high-speed package sorting or in monitoring stock in store shelves. A camera with Aitrios starts at $300 and is smaller than traditional cameras, does not include a fan or power cable and can run a large battery for a year. Aitrios works with Azure for basic setup of a system, but object recognition is at the edge and does not require the use of the cloud.