Impact case study database

Economic Impact

As a result of this multi-funded collaborative programme, Apical Ltd produced a prototype of the new device in 2016. Object detection algorithms in this prototype have been underpinned by fundamental research which has been later published in a series of papers [ R1]-[ R6] which included joint work with industrial collaborators [ R2, R3].

Fig. 1. Ingenic T01 video analysis AI coprocessor and Centrica Hive View Camera

In May 2016 Apical Ltd was acquired by ARM for USD350M, with the IPs for this technology (including the blueprints of the hardware and the object detection models). ARM then licensed the technology to Ingenic Semiconductors and Foxconn. Foxconn is the World's largest contract electronics manufacturer [ E5].

This led to the production of the T01 Spirit Professional video analysis AI coprocessor. The coprocessor is a core element of Centrica’s innovative Hive Smart camera, which can detect a presence of a person within a room helping to discriminate between alarms raised by pets which is not the case in cameras with generic motion sensors. The smart camera went on global market sale in January 2018 and in 2019 Centrica reported “ around 1.8 million customers now use Hive connected home products that can be controlled with just a tap on the app – from smart thermostats, plugs, lights and cameras, to contact and motion sensors” [ E9].

Through the project, ARM acquired and built new expertise and capabilities in Machine Learning and Computer Vision [ E4]. This directly impacted ARM’s ability to innovate new technologies “ that give increased performance and accuracy versus existing products” [ E4]. ARM is a leading developer of computational architectures and Intellectual Properties (IP) and is the keystone of the world’s largest computer ecosystem. “With more than 125 billion devices containing ARM-based chips, our architecture empowers innovation in multiple markets enabling partner innovation. ARM’s IPs feature in 90% of embedded devices globally.” [ E5]

“The research at the University of Leicester has underpinned the contribution to the development of the prototype System-on-Chip IP for a novel robust, high-throughput, accurate, and low-power and low-cost solution to the challenge of object detection in live video streams. …. The research also underpinned the development of ARM Object Detection Processor IPs.” [ E5]. As a result two patents have now been granted [US 10062013B2, US10489634B2].

ARM Object detection Processor IPs underpin ARM’s Project Trillium heterogeneous framework combining Cortex CPUs (100B chips as of 2017 [ E10]), Mali GPUs ( 6.8% of Global mobile phones including Galaxy S8 and S9 and out of about 5.5B mobile phones [ E8]), and Ethos NPUs (announced in October 2019) to deliver advanced machine learning (ML) user cases [ E7]. As a further evidence of economic impact, ARM created five graduate positions: “ As a result of our collaboration we now employ five postgraduate students from the Department of Mathematics, University of Leicester and are co-funding one of their joint PhD students.” [ E5]. All five students have been engaged in collaborative research with Apical which led directly to their job offers.

The project led to a new initiative, large-scale large-volume face recognition for security applications with VMS Ltd, a Scottish-based provider of security solutions with clients ranging from Ministry of Defence, Airports, Power Generation Plants to Museums and Art Galleries. At the centre of the initiative was the novel technology for detecting and processing faces from live video feeds and existing CCTV systems. The research “ has significantly contributed to almost a three-fold increase of the company’s turnover from about [GBP] 2M in 2016 to over [GBP] 7M in 2019 whilst also improving profitability”, and led to employment of 3 new graduate engineers and supporting 4 new graduate apprenticeships [ E6].

Understanding and Awareness

The underpinning research which has been developed as a result of collaboration between Apical and UoL and which was subsequently published in [ R1]-[ R6] informed Apical Ltd on the fundamental theory enabling the algorithm development [ E1- E3] and gave them “… insight into how to design future products that give increased performance and accuracy versus existing products. [ E4, E5]. “ARM customers want high accuracy object detection, tracking and recognition IP for a wide range of end applications – surveillance, automotive and healthcare to name but three. Non neural network approaches have the advantage of low cost but result in often unacceptable false positive rates. Neural networks have lower false positive rates but at a greatly higher cost. The KTP Project was able to validate that a combination of the two can yield the best of both.” [ E4]. The research also directly impacted on ARM’s capacity and investment plans: “ARM did not have active research in this field prior to the KTP Project”, “…. being actively engaged in research within these fields via KTP, ARM has been able to adjust our investment strategy to take advantage.” [ E4].

Finally, the research impacted on the preparedness of ARM for future business opportunities: “ARM Machine Learning Business Unit team have a much greater insight into emerging ML/AI/CV techniques than we did before the KTP project.” [ E4]