Page 11 - TT June for Digital

UPDATE

Bio-Inspired Camera Low carbon

rtificial intelligence innovation based on a different ‘bio bitumen’
combined with a novel bio- principle. Instead of a constant
inspired camera achieves frame rate, they have smart pixels research partnership
of technology and
100 times faster detection that record information every A
A of pedestrians and obstacles time they detect fast movements. industry experts led
than current automotive cameras. This Gehrig and Scaramuzza came up by carbon removal
important step for computer vision with a hybrid system that combines company CO2CO,
and AI and can greatly improve the the best of both worlds: It includes materials giant Tarmac,
safety of automotive systems and self- a standard camera that collects 20 sustainable materials innovation
driving cars images per second, a relatively low company Nanolyse Technologies
frame rate compared to the ones Ltd, Imperial College, London
Daniel Gehrig and Davide currently in use. Its images are and the University of Sheffield
Scaramuzza from the Department processed by an AI system, called is working on producing low
of Informatics at the University a convolutional neural network, carbon ‘bio bitumen’ made
of Zurich (UZH) have combined that is trained to recognize cars from algae. The team is trying
a novel bio-inspired camera with or pedestrians. The data from to use algae as a biomass source
AI to develop a system that can the event camera is coupled to a material to produce a black,
detect obstacles around a car much different type of AI system, called viscous, water-repellent material
quicker than current systems and an asynchronous graph neural that resembles petroleum-
using less computational power. network, which is particularly apt derived bitumen but offers a
Most current cameras are for analyzing 3-D data that change significant carbon reduction. The
early-stage research has shown
frame-based, meaning they take over time. Detections from the that bio-bitumen could be a
snapshots at regular intervals. Those event camera are used to anticipate viable alternative to the energy-
currently used for driver assistance detections by the standard camera intensive process of traditional
on cars typically capture 30 to 50 and also boost its performance. bitumen production.
frames per second and an artificial The result is a visual detector that
neural network can be trained to can detect objects just as quickly
recognize objects in their images as a standard camera taking 5,000
-- pedestrians, bikes, and other cars. images per second would do but
requires the same bandwidth
Combining the best of two as a standard 50-frame-per-second
camera types with AI camera.
Event cameras are a recent Source: ScienceDaily TT

www.trafficinfratech.com / June 2024 / TRAFFICINFRATECH 11