Accelerating innovation: transforming businesses through smart sensing and AI
Chris Shore
Offline Chris Shore
8 days ago
The way we interact with technology is changing through the power of artificial intelligence (AI) and machine learning (ML). New ML platforms such as voice assistants and smart wearables are revolutionizing how people and technology interact. ML models can train data to create specific instincts for use in all kinds of intelligent products – from fall detection clothing to predictive maintenance tools.
Chris Shore, Director of Embedded Solutions at Arm, recently sat down with Imagimob’s CEO and Co-founder Anders Hardebring to learn more about their intelligent software solutions that are empowering customers with actionable insights in real-time.
Imagimob’s story
Imagimob is a software company founded in 2013 by Anders Hardebring, Tony Hartley and Alexander Samuelsson. Their aim is to help customers develop innovative, smart applications through on-device or edge processing. Imagimob’s mission is to empower customers with actionable insights that can help them make informed decisions about their products. And from the start they had a strong focus on achieving high accuracy while being resource effective in terms of memory and power consumption.
Imagimob’s customers comprise of large organisations, such as Scania, Husqvarna and ABB, and the applications Imagimob support is endless! From personal safety devices to handheld power tools and advanced sensors for predictive maintenance of electric drives – the possibilities of Imagimob’s software solutions are wide-ranging. More recently, the company has been involved in a research project where its edge AI technology is being used to improve productivity in the agriculture sector. In this project, the aim was to minimize the use of antibiotics in farming by introducing active monitoring of live crops using Edge AI technologies. This farming supports the wellbeing of both animals and humans.
Imagimob Q&A
Can you tell us more about one of the projects you’re currently working on?
We have an exciting project in the works with Scania’s groundbreaking C-me Safety Vest, which hits the market this year. The intelligent and highly visible vest combines Imagimob Edge AI software with a unique active light technology system by LightFlex, the unique active light technology in the C-me Vest. The Vest uses the Arm Cortex-M4 processor with 32 kB of RAM due to its superior performance at good value, as well as its low power consumption.
The connected and intelligent C-me Safety Vest keeps a protective hand over truckdrivers working around vehicles. A truck driver is one of the most dangerous professions in Sweden, according to official statistics. In the last 10 years, 59 truck drivers have died in Sweden from traffic accidents and accidents during loading and unloading, on dark winter roads, in narrow city streets and in dark parking spaces. If a driver is to fall or become immobilized suddenly, the C-me Safety Vest automatically triggers for help. Pre-assigned contacts are notified and they are automatically sent GPS coordinates to the point of contact. Also, the flashing light indicator is switched on instantly as the call for help is triggered. By using data from embedded sensors, Imagimob AI is instantaneously able to analyze the motion of the driver. The result is a considerable safety boost for drivers who typically work alone in high-risk environments.
The intelligent C-me Safety Vest
The intelligent C-me Safety Vest is an excellent example of how we’ve used Arm technology to empower and equip our partners. When discussing the Arm-based technology used in our products with our customers, they instantly understand the capabilities and opportunities – it’s an easy discussion and everyone is on the same page when using Arm technology.
I read about your impressive benchmarking results for Imagimob AI against deep learning. Can you tell us a bit more about this project?
Certainly! In the last year, we have been working on a benchmark study that compares Imagimob AI technology against a state-of-the-art deep learning network for time series classification. While the results revealed a similar level of accuracy between the two systems, Imagimob AI proved to be significantly more resource-effective when it came to RAM memory usage and CPU operations.
The results showed that the Imagimob AI system was 2% more accurate than the deep learning model. A difference that, while minor, is very impressive when considering the drastically different amount of computational resources used by the two approaches. Compared to Imagimob AI, the deep learning model used 33 times more RAM memory and required 800 times more instructions to make a single classification. By using Imagimob AI, customers can use less computing power and therefore, a lower bill of materials and the battery will last longer. Also, Imagimob AI allows customers to fit more AI models, applications and functions on the same chip.
Imagimob AI is a mixture of algorithms inspired from decision tree based algorithms to neural network based approaches. Imagimob have developed an extremely resource-efficient network compiler that can use weight architectures (e.g. from high-level neural networks API’s such as Keras) that are compressed into models. This can easily be deployed in resource-constrained environments, such as on the Arm Cortex-M processor family.
One of our internally developed algorithms is bitEnsemble, which is a random forest ensemble method working on bit strings. The benchmarking report is comparing bitEnsemble and a standard deep learning approach using LSTM:s on time series data.
The Imagimob AI system used in the benchmarking report uses around ten percent of the available memory on an Arm Cortex-M0 processor and requires only ten milliseconds per classification. This is a crucial component that enables Imagimob AI to run efficiently on small devices, while still delivering a high standard of performance and accuracy.
If you’d like to learn more about the benchmarking study, details can be found in this white paper.
It’s not just about efficiency - the great thing about Arm Cortex-M processors is that the portfolio continues to evolve and expand, providing greater potential for our products in the future. That’s one of the reasons we will continue to use Arm-based technology at Imagimob. We’re very excited for new developments in the machine learning space, especially the new architecture extension Arm Helium technology for future Cortex-M based processors – this is set to bring up to 15 times performance uplift for ML, which will take our product to the next level in terms of the data we can process and the insights we can deliver.
What do you think is going to be the most significant change in the electronics industry in the next five years?
Changing the focus on cloud AI to edge AI opens great possibilities in several areas for us. Applications ranging from intelligent low powered IoT devices to fully autonomous systems that operate independently of a cloud connection will undoubtedly change for the better through improved performance and efficiency.
Also, solutions for efficient powering of devices using energy harvesting will make battery powered systems look like dinosaurs! The shift to self-powered devices will also drive power efficiency for MCU development and edge computing.
What challenges will that change bring for your company and your customers?
For Imagimob and our customers, more intelligence at the edge will open a new world of possibilities. We are working in a couple of new, exciting projects with Husqvarna, a leading supplier of power tools. These projects were not possible due to hardware cost and power constraints.
Learn more about Imagimob by visiting their website or by contacting Anders Hardebring more information.
Accelerating innovation: transforming businesses through smart sensing and AI
The way we interact with technology is changing through the power of artificial intelligence (AI) and machine learning (ML). New ML platforms such as voice assistants and smart wearables are revolutionizing how people and technology interact. ML models can train data to create specific instincts for use in all kinds of intelligent products – from fall detection clothing to predictive maintenance tools.
Latest in Home