“This will become the world's first deep learning-enabled roadside advertising system with vehicle specific, real-time targeting. One major outcome of the project is demonstrating deep learning algorithms utilizing HyperStore's sophisticated meta-data to realize real-time recognition and classification. Big data that otherwise might have remained stagnant can be automatically turned into smart data, and highly efficient analysis can be provided," said Michael Tso, CEO and co-founder, Cloudian. “Cloudian's smart data storage was built on three principles: 100 percent native S3 APIs, hybrid cloud, and tight integration of analytics with storage. HyperStore’s smart storage capabilities, which can be further enhanced by deep learning, rapidly and cost effectively turn data into actionable business insights and revenue opportunities. We look forward to deepening our partnerships to expand deep learning into transformational use cases in retail, healthcare, genomics, and IoT."
Cloudian kicked off the project by providing the HyperStore software with training data that consisted of a large volume of vehicle information, images and video of car models, plus vehicle attribute inputs. This information was classified using HyperStore’s smart data storage functionality and will be tested to accurately identify vehicle models on Tokyo roadways. As part of this experiment, HyperStore will also capture detailed, real-time data related to traffic volume at various times in the day, which can be made available to public institutions such as the Ministry of Land, Infrastructure and Tourism, local municipalities in Japan and to enterprises for retail location planning.
“In today’s fast-moving advertising world where consumers expect tailored experiences and uniquely relevant messages, we are thrilled to work with Cloudian and QCT to pilot an innovative, highly targeted advertising approach,” Ichiro T. Jinnai, director, Out Of Home Media Service Division, Dentsu, Inc. “We will bring our proven global advertising expertise to bear for planning, sales and media development of the ads that are automatically served based on the deep learning analysis from HyperStore about the approaching vehicle. Cloudian is the perfect partner to lead this effort and showcase its smart data storage capabilities for a new era of IoT experiences.”
An aim of the project is to apply the automated vehicle recognition to generate targeted display advertisements based on vehicle model; for instance, an eco-friendly product could be displayed to drivers of hybrid/electric vehicles. Large LED billboards will be used in this portion of the experiment. The system neither captures nor stores identifiable vehicle information, including license plates. While specific advertisers have not yet been identified, a recent press announcement in Japan has resulted in a number of inquiries to the participating companies.
The project also plans other demonstration experiments of new real-time advertising based on the analysis of not only vehicles but also human behaviors, such as attributes matching ads at shopping malls and tourists sites.
“Cloudian’s deep learning solution, powered by the Intel® Xeon® processor E5 family on Quanta’s servers, offers the flexibility and performance to deliver a high-impact experience for smart, connected cities,” said Ron Kasabian, vice president, Big Data Analytics, Intel. “We are pleased to support Cloudian’s deep learning work using Caffe optimized for Intel Architecture, as this solution will benefit customers that do infrastructure planning, business development, healthcare, digital advertising and more.”
"QCT is delighted to provide scale-out storage and compute to support this project," said Mike Yang, President of QCT. "Cloudian HyperStore smart object storage tightly couples analytics and storage, delivering unparalleled efficiencies for new use cases like responsive advertising and intelligent cities. I look forward to expanding this partnership and propelling QCT's renowned hyperscale infrastructures to the next frontier."