Skip to Main content Skip to Navigation

Techniques de spatialisation binaurale pour le guidage de sportifs non-voyants

Abstract : In continuity with other work of the binaural community we believe that spatialized sound can be an effective tool for guiding blind people, including for sports practice.A system for binaural audio guidance must be sufficiently accurate and responsive to be able to take into account every movement of the subject. This has led us to the development and implementation of a real-time localization system and a low-latency binaural spatialization software. Finally, we integrated the whole system into an embedded device.The advanced global satellite navigation techniques (augmented and multi-band) are not always available indoor or in urban environments. This is why we have been working on alternative methods of location and real-time tracking for indoor applications. First of all we have developed a robust real-time calibration and lateration method using an Ultra WideBand beacon network using a Kalman filter (UKF). We have also developed an original localization method using an unmodulated continuous wave Doppler radar array. We showed that it is possible to use the amplitude of the Doppler signal to estimate the distance to a moving object. We have then implemented a particle filter to allow real time localization by hybridization of distance data, measurements of radial Doppler velocity augmented by the heading provided by an inertial unit.In the field of acoustics and binaural audio, we have focused on understanding people's abilities to locate and track sound in motion. To do this we conducted experiments using natural and spatialized sounds using a binaural audio engine. These experiments showed that, on the azimuthal plane, spatialized audio stimuli allow for a localization comparable to natural sounds, even with non-individualized and interpolated HRTFs (head-related transfer function). Moreover, we were able to show that even in the azimuthal plane, the stimuli obtained by HRTF convolution are superior to panning (ITD+ILD) both for fixed and moving sounds. Based on the previous work of the team, we have implemented efficient algorithms for the real-time sound spatialization on embedded platforms with limited resources. For an efficient implementation, this real-time approach implied a thorough understanding of the sources of latency whether they are related to head-tracking or to the audio subsystem of modern operating systems.Finally, we used these localization methods and audio techniques to build a guidance device where a sound source continuously precedes the person to indicate the path to follow. It has been designed in collaboration with visually impaired people, in an iterative process and with an approach focused on the user needs. We then carried out guiding experiments with blind people in conjunction with our associative partners, which made it possible to evaluate different control strategies. We have thus been able to confirm that spatialized sound could constitute an effective tool for guiding blind people, without inducing any penalizing cognitive load, for practicing sports such as walking, running or rollerblading in partial autonomy, including the context of performance.
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Friday, July 23, 2021 - 11:33:10 AM
Last modification on : Saturday, July 24, 2021 - 3:47:20 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03297189, version 1



Sylvain Ferrand. Techniques de spatialisation binaurale pour le guidage de sportifs non-voyants. Traitement des images [eess.IV]. Institut Polytechnique de Paris, 2021. Français. ⟨NNT : 2021IPPAX026⟩. ⟨tel-03297189⟩



Record views


Files downloads