Pyramics, Inforce Computing, Fraunhofer IIS, and Basler come together to help implement a state-of-the-art edge-compute IoT system for a retail video analytics application.
The digital transformations taking shape today in several industries will bring extremely profound changes to the way we will live our lives or how companies operate. The heady mix of people, less-expensive sensors, actuators, compute, connectivity, and big-data analytics has indeed made internet-of-things (IoT) or internet-of-everything (IoE) a big living reality. While one might debate the cost of value propositions these transformations bring to end-users at large and who will pay for it, one can’t help but recognize that the benefits are starting to look pervasive.
Video analytics in the retailing and out-of-home media markets
There are a number of areas where disruptive technologies and business models will harness new opportunities and bring excellent outcomes for both the end-user and the companies that provide innovative solutions. One such set of examples is the transformation out-of-home advertising, market research, and in-store retailing industries are about to witness. We live in a data-driven world, so higher the accuracy of the acquired data, better/faster the insights we can derive from it, and lower the costs of getting to meaningful and actionable insights. Better insights means a more satisfying shopping experience for the consumers and the retailers.
Capturing data about age, gender, ad interactions and subsequently observed emotions of the audience may provide a better means to understand customer preferences and shopping behavior. This can be used to tailor personalized marketing campaigns to match customer needs, bringing in an improved RoI. Things to measure may include:
- Advertising reach: Measure # of folks passing by an ad display and # that looks at ads.
- Advertising effectiveness: Measure age, gender, dwell time, ad interactions, emotions
- Demographic people counting: When do particular groups visit?
- Conversion rates at PoS: Measure revenue distribution based on shopper demographic
Capturing high-quality video under limited and varied lighting conditions can be a huge challenge. The ability of compute-intensive facial recognition/analysis algorithms to accurately measure the above set of data becomes critical under such circumstances. While the cloud may offer cheaper and infinite compute resources, current European e-privacy laws, for example, prohibit the uploading of any personally identifiable information to the cloud.
Secondly, meaningful and actionable insights have to be realized in real-time or near-real-time. This means that the data has to be close to the compute—any latency associated with uploading of data to the cloud for processing is not acceptable.
Thirdly, uploading terabytes of streaming video to the cloud requires expensive and high-bandwidth connectivity. As a result, it makes a lot of sense to run the facial recognition algorithms locally before uploading only anonymized data (metadata) such as age, gender and emotion for further analytics in the cloud. This requires high-performance compute at the edge of the IoT network. This could also ease bandwidth usage globally and bring down the total cost of ownership. Here’s a story of how such a system was rapidly put in place recently.
Berlin, Germany based Pyramics® has designed a cool and advanced video analytics device powered by the Inforce 6501™ Micro SOM (Qualcomm® Snapdragon™ 805 based). Four diverse parties got together in a very short amount of time to help design the system, including a small form-factor custom carrier board, interfacing a USB3.0 camera with device drivers for Linux OS, and run an advanced facial recognition software application to provide real-time actionable analytics. This culminated in a great demo that was showcased at the Embedded World Conference in Nuremberg, Germany, February 2016. The results were also published recently at the EuroCIS 2016 conference in Dusseldorf, February 2016.
The four parties….
Pyramics provides the data and hardware solutions to measure the success of various video analytics based marketing efforts. Pyramics’ customers are in the business of out-of-home advertising, market research, and retailing. For this purpose Pyramics developed an intelligent optical sensor, Pysense, in partnership with Fraunhofer Institute for Integrated Circuits (facial recognition software), Basler (high resolution cameras), and Inforce Computing (Snapdragon based compute platforms). Pysenses capture anonymous data (metadata) about the shopping behavior and ad interactions of customers at the point-of-sale (PoS). This Information is then used to optimize marketing strategies.
So, what’s really under the hood?
With a tight 8-week turn-around window to build completely working prototypes, Inforce assited Pyramics in rapidly designing a custom carrier board to fit the enclosure requirements of the Pysense, which can also be powered by an optional Power-over-ethernet (PoE) port. This included a design assistance services package from Inforce that provided direct technical support and schematics of a reference carrier card. Inforce worked closely with Pyramics and Basler while the Fraunhoffer ISS’ world-leading SHORE™ software for facial recognition was ported to the Linaro Linux OS running on the Inforce 6501 Micro SOM. For the high-resolution video capture, device drivers to enable the Basler dart series daA2500-14uc USB3.0 camera to interface with the Inforce 6501 Micro SOM were developed. Bringing it all together, the system was demonstrated at the Embedded World conference in Nuremberg, February, 2016 and was a big hit with the visitors.
|Compute Module||Inforce 6501 Micro SOM + Custom carrier board|
|Processor||Qualcomm Snapdragon 805 (APQ 8084)|
|Connectivity||Onboard GbE (WLAN), BT 4.1, GPS, UMTS/LTE (external)|
|Memory||2GB LPDDR3 and 16GB eMMC flash|
|External interfaces||Micro USB3.0, Micro HDMI, SD-card connector,|
|Dimensions||34mm x 105mm x 26mm|
|Power Supply||230v / Power-over-Ethernet (PoE)|
|Operating System||Linaro Ubuntu Linux 15.X|
There are several contemporary IoT use-cases such as retail video analytics, where a higher degree of compute is required at the edge. The continued lower cost of mobile technology based high-performance compute coupled with significantly reduced power consumption will spawn a new generation of IoT use-cases that provide reduced latency to actionable insights, better control over privacy and security, and scalability in large deployments.
Inforce Computing Desk
Images credit: Pyramics, Basler, Inforce Computing