The concept of UWB radar sensor - TheBlue.ai

Blog. Immerse yourself in AI

Marianna_Blog_Photo_fin

Marianna Parzych,
Image Processing Engineer
2 June 2020

The concept of UWB radar sensor

The key concept of the radar sensors is to send out signals and then receive signals reflected by objects. The IR-UWB sensor (pulse radar ultra-wideband) – as we use it – sends short-term pulses, whereby the duration of a single pulse is usually in the order of a few nanoseconds to a few hundred picoseconds [1]. Such a signal has a number of desirable properties such as fine resolution, good penetration, multi-path immunity. IR-UWB radar sensors are extremely precise and use little electricity.

Occupancy detection - how did we do it?

In addition to the desired signal of a person, the received signal also contains information about static objects in the environment such as furniture, wall, chair or desk. It was necessary to suppress the interference signal. A simple running averaging filter [3] was used to generate the clutter map. After subtracting the clutter map, all reflections in the signal were associated with moving targets. However, the signal was still quite noisy. For a simple application, namely occupancy detection, this was not a problem. Nevertheless, it would e.g. in the case of people counting noises, it can be very difficult to recognize every person. Simple filtering methods like Gaussian or median filters could not help us – they removed the reflections of moving targets together with the noise. We opted for more sophisticated wavelet filtering [4]. Then we looked for the greatest reflection in the picture. If it had a value greater than a chosen threshold, we accepted it as a moving object – a person – within the radar frame. The position of the reflection was then used to calculate the distance to the person being detected, assuming that the resolution of the received signal was 5.14 cm [5].

How do you implement a radar-based system with AI?

RF data received from the sensor is a set of consecutive 1-dimensional signal frames. Each frame corresponds to reflections at a discrete moment, while events such as traps or fainting spells last longer. As a solution, we suggest stacking the sequence of consecutive frames into a 2-dimensional matrix that contains information about a selected time period. The advantage of this approach is that the resulting matrix can be treated like an image so that we can take advantage of the Convolutional Neural Networks (CNNs) [6].

With Edge AI devices like NVIDIA Jetson, we can use our neural network development and training expertise to implement a solution tailored to your needs. By using the edge AI devices, calculations can be carried out anywhere. This means you can use radar-based AI solutions for intelligent buildings, security, healthcare or retail without having to transfer data. This means full respect for user privacy and full compliance with the GDPR.

1 Xuba Wang, Anh Dinh and Daniel Teng, Radar Sensing Using Ultra Wideband – Design and Implementation, 2012, http://dx.doi.org/10.5772/48587

2 X4 – Datasheet , Xethru Datasheet by Novelda, 2020

3 Piccardi M., Background subtraction techniques: a review in Systems, Man and Cybernetics, 2004

4 Burrus, C. S. and R. A. Gopinath, H. Guo. INTRODUCTION TO WAVELETS AND WAVELET TRANSFORMS, A PRIMER. Upper Saddle River, NJ (USA): Prentice Hall, 1998

5 X4M300 Datasheet, Xethru Datasheet by Novelda, 2017

6 Xiuzhu Yang, Wenfeng Yin, Lin Zhang, People Counting Based on CNN Using IR-UWB Radar, 2017

You have an idea on how AI can support your business or you have a question about the UWB radar sensor? – Contact us.

We look forward to hearing from you.