Outdoor Group Counting Based on Micro-Doppler Signatures Obtained with a 77GHz FMCW Radar

• Published in: 2024 21st European Radar Conference (EuRAD) pp. 376 - 379
• Main Authors: Cakoni, Dejvi, Storrer, Laurent, Cornelis, Bruno, De Doncker, Philippe, Horlin, Francois
• Format: Conference Proceeding
• Language: English
• Published: European Microwave Association (EuMA) 25.09.2024
• Subjects: Accuracy; CNN; Convolutional neural networks; FMCW radar; micro-Doppler signature; Pedestrians; people counting; Radar; Radar applications; Synthetic data; System performance; Time-frequency analysis; Training; Transfer learning
• DOI: 10.23919/EuRAD61604.2024.10734953

In numerous mass gathering settings along with daily commutes, maintaining an accurate count of individuals is imperative. Radar systems, known for their cost-effectiveness and low energy consumption, facilitate discreet monitoring across various applications. In this work, data was collected via a 77GHz frequency-modulated continuous wave radar (FMCW) in an outdoor pedestrian street. We leverage the unique gait model of each individual, which results in a distinct instantaneous velocity pattern as a function of time to be able to count people. Therefore, we analyze and process our data in the time-frequency domain to generate the so called micro-Doppler signatures (MDS). Then, these MDS are fed to a Convolutional Neural Network (CNN) to classify groups of different sizes. Furthermore, due to the lack of significant amount of data, the CNN was firstly trained with synthetic data and later on with the measurement data, to increase the system performance. The proposed system overcomes the limitations of existing camera-based people counting techniques such as being affected by lighting conditions and distinctly from other radar related work, targets an outdoor scenario.