Enhancing People Counting in Cluttered Environments Using mm-Wave Radar, LSTM, and Ensemble Learning

Radar-based sensing in cluttered environments to ensure safety, and optimize resource allocation is a challenging task. Issues such as unwanted reflections, target signal blocking, and the need to collect new datasets for each new environment make it an effort-intensive process. To tackle these challenges, we’ve developed a radar-based system for counting people, making use of our previously developed Orthogonal Time Frequency Space joint communication and sensing system. To make sure that the received radar signals are as clear and as useful as they can, we performed several required preprocessing steps, such as static clutter removal or wavelet thresholding. We employed both CNN and LSTM networks to capture the spatial and temporal patterns in the training data from two different environments. To make the system more reliable and to make it able to generalize to new, unseen industrial settings, we incorporated an ensemble learning method with a voting mechanism. The effectiveness of this approach is promising in that it achieved 85% accuracy in counting up to 3 people in dynamic and cluttered environments. This shows its potential to effectively deal with the key characteristics of real-world people counting tasks.