Imagimob Combines its AI Software Toolchain with IAR Embedded Workbench to Boost Safety and Performance in Edge AI Applications

15 October 2019

For many companies seeking system certification for Edge AI applications, the development tools they use are an important part of the equation. Functional safety has become one of the most important features in embedded systems—especially within the automotive, railway, and medical market segments.
 
At Imagimob, our specialty is artificial intelligence products for edge devices, so this demand for functional safety is definitely something we have noticed among our customer base. That’s why, when the opportunity arose to conduct a test with the toolchain IAR  Embedded Workbench—we seized it.
 
The Test
The Imagimob AI Software Suite is a set of software tools for Edge AI application development that enables small devices with intelligence and data processing capabilities.
 
IAR Embedded Workbench is a compiler and debugging toolchain that provides a complete IDE, helping to ensure quality, reliability and efficiency in embedded applications.
 
We wanted to see if, by combining our seemingly complementary tools in the same development workflow, we could manage to boost the functional safety and performance of an Edge AI application.
 
So, we conducted a benchmark test.
 
Step 1
Imagimob AI was used to collect and label data and to generate and verify an optimized AI model. The AI model was a mix of Convolutional Neural Network layers (CNN), Long-Short-Term-Memory layers (LSTM) and fully connected network layers (FCN). In total, the system consisted of eight layers.
 
Step 2
The final output from Imagimob AI was generated as C-code, optimized to run on an MCU with constraints on memory and execution speed.
 
Step 3
The IAR Embedded Workbench was used to compile and further optimize the generated C-code for an STM32WB55 Nucleo board, targeting an ARM M4 with a floating point processor.
 
The Results
The final application was successfully deployed on the target platform, performing one forward pass (prediction) in just 0.28 milliseconds.
 
That is to say that the benchmarking test demonstrated a successful integration of Imagimob AI and IAR Embedded Workbench throughout the entire development chain of an Edge AI application.  
 
“We are really excited about the results,” says Anders Hardebring, CEO and Co-Founder of Imagimob. “There’s great potential here to create an application that not only delivers on performance, but also ensures the functional safety and support many of our customers are asking for.”