[cs-talks] IVC Reading Group, Andrea Zunino, 12/9 at 3pm in MCS 148
Harrington, Jacob Walter
jwharrin at bu.edu
Mon Dec 5 15:52:40 EST 2016
Intention Prediction: A New Paradigm in Action Prediction and
Novel Audio Tracking by Means of Spectral Signature
Andrea Zunino, Third Year PhD Student, Istituto Italiano di Tecnologia, Genoa
IVC Reading Group, Friday December 9th from 3pm – 4pm in MCS 148
Abstract: In the first part of my talk I would like to present my actual research activity (intention prediction): Previous attempts in classifying future or unfinished actions utilize developing motion patterns which are specific of the subsequent actions, since they contain some cues that undoubtedly help the recognition. Further, another important aspect of the entire activity recognition problem is that the current techniques are mainly exploiting the scene context to support the classification. However, although this information could help, it can be insufficient to solve the task or, worse, the context may not always be available or easily recognizable, being also misleading when the scene is too noisy or cluttered.
In any case, an important source of information to disambiguate intentions can be provided by the kinematics of the movement. In fact, recent findings from behavioural neuroscience indicate that how a motor act is performed (e.g., grasping an object) is not solely determined by biomechanical constraints imposed by the object’s extrinsic and intrinsic properties with which one is interacting but it depends on the agent’s intention. As a new paradigm, we aim at introducing a brand new challenging action prediction problem consisting in the prediction of human intentions, defined as the overarching goal embedded in an action sequence.
In the second part of my talk I would like to present my previous works, mainly regarding my master thesis: We proposed a novel method for generic target tracking by audio measurements from a microphone array. To cope with noisy environments characterized by persistent and high energy interfering sources, a classification map based on spectral signatures is calculated by means of a machine learning algorithm. Next, the classification map is combined with the acoustic map, describing the spatial distribution of sound energy in an environment, in order to obtain a cleaned joint map in which contributions from the disturbing sources are removed. A likelihood function is derived from this map and fed to a particle filter yielding the target location estimation on the acoustic image.
Bio: Andrea Zunino was born in Ovada (Italy) in 1989. He received his Master's Degree in Multimedia Signal Processing and Telecommunication Networks (MSPTN) in 2014, with a thesis about tracking of sound sources through an acoustic camera device developed in Pattern Analysis and Computer Vision (PAVIS<https://www.iit.it/it/research/lines/pattern-analysis-and-computer-vision>) department, at Istituto Italiano di Tecnologia (IIT<https://www.iit.it/it/>) in Genoa. Currently, he is a third-year fellow PhD student at the Istituto Italiano di Tecnologia, PAVIS department under the supervision of Prof. Vittorio Murino. His research activities are related to audio/video signals processing, activity recognition/prediction, for video-surveillance and psychological intention-understanding.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks