Validation of predictive algorithms for the estimation of the number of people in a canteen

This study aims to estimate the number of people in a canteen from the babble noise level in the room. Noise levels were measured in the CIRCOOP canteen of the Politecnico di Torino across 4 days during the COVID pandemic, while three people counters, based on IR sensors, were located at the entrance and at the exit of the canteen. Reverberation time was also measured to calibrate the acoustic model in Odeon 16 and Grasshopper application of Rhinoceros 7 was used to calculate some parameters needed for the application of two prediction algorithms. The former assumes a diffuse field while the latter does not, and instead it considers the rate of spatial decay per distance doubling and the interpersonal distance. Besides the acoustical parameters of the room, the models need as input the group size g and the Lombard slope c, which strongly depend on human context. In the case of this canteen, the best matching was obtained with g=8 and c=0.5 for both the models. Our results showed that the prediction of the number of people from the babble noise is possible only for noise levels lower than 70 dB(A).