AI based smart monitoring system on an embedded system
Description of the company NXP Semiconductors enables secure connections for a smarter world, advancing solutions that make lives easier, better, and safer. As the world leader in secure connectivity solutions for embedded applications, NXP is driving innovation in the automotive, industrial & IoT, mobile, and communication infrastructure markets.
Situation Nowadays, artificial intelligence (AI) is used to solve a wide range of problems in a wide variety of application areas. AI analysis in the cloud is not always desirable due to security, data rates, latency, or availability.
With AI at the edge, our devices-from smart thermostats to autonomous cars-rely on patterns and inference to learn, adapt, and make decisions in real time without the latency and bandwidth challenges introduced by the cloud.
Problem One of the challenges AI poses is their large computational complexity and the massive parameter count (e.g., ResNet50 requires over 23M parameters and 4 GFLOPS per frame). Therefore, AI inference is relatively slow and power-hungry on traditional general-purpose compute architectures. Especially on embedded devices the AI inference is challenging. To address this issue optimizations techniques such as integer quantization and pruning were designed, enabling seamless AI employment at the edge.
Aims of the project A demonstrator is planned and built to monitor a specific application. The aim of the project is to develop a model for the detection of anomalies. The developed model will run on an embedded system and must be optimized for this to be able to inference based on AI given the limited computational capacity.
For the investigation of an application specific AI on an embedded system several scopes need to be covered.
Setup of the sensor system and electronics
Research on state-of-the-art AI algorithms and optimization for embedded systems
Develop, train, and optimize the model for an edge device
Visualize the AI output with a graphical user interface
Summarize and present the results
Target group (students) Students with a background in engineering, signal processing, software engineering, computer vision, or similar.