Pages:     | 1 |   ...   | 55 | 56 || 58 |

To extract repeatable behavioural Every frame undergoes several patterns, a training set of images and videos routines: pilots face detecting (using Violadepicting persons activity is required. In order Jones algorithm [5]), identifying the state of an to produce such a data set, seven experiments eye open or closed (by nearest neighbors were carried out in the aircraft TU-154 pilot's classifier using histogram of gradients cabin simulator with an option of description of eyes area [6]). After analyzing extraordinary situations modeling (for the sequent frame, the following values are example, weather conditions changing, engine recorded in the log-file: face detection results failure). Above the first pilot's dashboard a (1 found, 0 not found), face coordinates (in color web-camera Logitech c910 equipped pixels), face size (in pixels), blinking detection with a stereo microphone was mounted.

result (0 eyes opened, 1 eyes closed).

Recorded video resolution was 640x480 px.

Before the beginning of the Results experiment, the adjustment procedure was conducted. To estimate the extreme points of Log files containing face coordinates the persons head position, the pilot was asked and blinks data were processed to find out to look over the dashboard. To create template pilots repeatable behavioural patterns and images of the person with their eyes closed or estimate their motor activity trends. Face open, the correspondent pictures were taken.

capture percentage was the first evaluated In order to monitor long-term changes variable. It demonstrates reliability of the face of behavioural characteristics in different grabber. The mean of the face capture flight conditions, the experiment was divided percentage during the long flights experiments into two parts: long monotonous flight without (n1 = 4) was 76% (the minimum value 46%, extraordinary situations; short flight with the maximum value 96%). The mean of the emergency situations (engine failure or strong face capture percentage during the short flights side wind when landing). The duration of the experiments (n2 = 3) was 84% (the minimum first period was approximately 60-90 minutes, value 54%, the maximum value 99%). Due of the second period 8-20 minutes.

to insufficient lighting, wide range of pose EEG and EOG (electrooculogram) variations, occlusions, caused by EOG electrodes were applied to the pilot before the electrodes and glasses, this percentage varies.

flights. Tapping test and reaction test were The low precision in blinking detection was taken before, after and between the flights.

observed by the same reasons. Therefore, pose After the preparations are over the variations such as face size and coordinates flights begin. The following data are recorded:

were the most informative behavioural signals video of the pilot, pilots EEG signals and in this series of experiments.

flight protocols, which contain information Pilots pose dynamics within long and about pilots and dispatchers actions and short flights was examined to find out emergency situations.

repeatable behavioural patterns. The most Due to high-frequency noise in the stable patterns were detected in the face size video-signal, measuring low-amplitude dynamics, which depends on the distance parameters such as size of an eyelid cleft is between the face and the camera. Two impossible. Therefore, low-frequency, highexamples of the face size variations during amplitude motions such as head pose aircraft takeoff and climbing are shown in the variations are the most informative features in Fig. 2. There is a specific trend of the pilots this video stream.

moving away from the camera.

XVI Another repeatable behavioural trend The human motion analysis techniques of the pilot moving towards the camera was for functional state estimation were described.

detected while landing (Fig. 3). The techniques were tested within the task of pilots functional state estimation. The standard deviation of face horizontal coordinate was analyzed as the attribute of motion activity intensity. Repeatable motion patterns and significant contrasts between motion parameters in the different parts of the flight were detected. In particular, the highest level of pilots motor activity was detected during landing, the lowest was observed Figure 2. Two examples of the face size dynamics after climbing. Deviations from repeatable during takeoff and climbing of the plane.

behavioural patterns were considered as indicators of unusual and possibly non-optimal functional state.

Acknowledgments The work is supported the Russian Foundation for Humanities, grant 11-06-00704a and Russian Foundation for Basic Research, grant Figure 3. Two examples of the face size dynamics 11-01-00750a.

during landing of the plane.

References 1. Adrian J. Xavier. Managing human factors in The flight was divided into four parts aircraft maintenance through a performance excellence in proportions: 20% of flight duration (takeoff framework // A Graduate Research Project, Embryand the beginning of the flight), 30% the first Riddle Aeronautical University, 2005.

part of the flight, 30% the second part of the 2. Lawrence Barr, Heidi Howarth, Stephen Popkin, flight, 20% the end of the flight and landing.

Robert J. Carroll. A review and evaluation of emerging The behavioural trends within these intervals driver fatigue detection measures and technologies. // A were compared with each other. Face Report of US department of transportation Washington horizontal coordinate standard deviation was DC, 2005.

considered as an integral indicator of motor 3. Robert Schleicher, Niels Galley, activity. The mean of this value within the first Susanne Briest, Lars Galley. Blinks and saccades as indicators of fatigue in sleepiness warnings: looking parts of the flights was 15.72.4 px, the tired // Ergonomics, 2008, 51(7):982-1001.

seconds 9.32.3 px, the thirds 11.32.8 px, 4. Roman Bittner, Pavel Smrcka, Miroslav Pavelka, the fourths 19.52 px. According to this Petr Vysok, Lubomir Pousek. Fatigue indicators of results face pose standard deviation was the drowsy drivers based on analysis of physiological lowest within the second stage of the flight.

signals. // Lecture Notes in Computer Science, 2001, Therefore, motor activity level was the lowest.


The fourth flight part (landing) was the most 5. Paul A. Viola, Michael J. Jones. Robust Real-Time active stage.

Face Detection. // International journal of computer The face pose standard deviation over vision, 2004, 57(2):137-154.

the long flights was 18.30.6 px, the short 6. M. Petrushan, Y. Vermenko, D. Shaposhnikov, S.

flights 152.6 px. Thus, the difference Anishchenko. Analysis of colour- and grayscale-based feature description for image matching // Proceedings of between motor activity intensities over the 8th Open German/Russian Workshop on Pattern long and the short flights was insignificant.

Recognition and Image Understanding, 2011, OGRW8-2011:237-239.

Conclusion 4- ۻ CONTINUOUS ATTRACTOR MODEL OF SIGNAL SPATIAL PROCESSING PROCEEDED BY GRID CELL Z.S. Yeremenko, V.D. Tsukerman, A.A. Sazykin, S.V. Kulakov A.B.Kogan Research Institute for Neurocybernetics, Southern Federal University har_zs@rambler.ru The report presents the results of the investigation of the result of network architecture performing continuous attractor model of spatial environmental special spatial cognition functions.

signal coding, intended for navigational task solving.

In particular, its not clear how spatial This investigation shows formation of neural information about direction and place is associations on different levels, encoding of the basic represented on neural network level and how spatial variables place, direction, linear and angular velocity, as well as firing coordination of neural these basic variables are processed in assemblies in navigational behavior.

hippocampal formation.

In the report it would be shown how Introduction functional specialization and clusterization of neural assemblies solving spatial processing The last data concerning specific cells of tasks occur in the same continuous attractor hippocampus, entorhinal and parietal cortex, ECI-network (even cyclic inhibitory network).

determining spatial cognition, memory and Also it would be shown how the most navigational behavior of highly organized important spatial variables, place and direction animals and humans, raised the whole range of are encoded and how neural correlates of hypotheses and computational models, selflocalization are determined.

accounting for the formation of place cells Suggested model has principle differences [1,2], head direction cells [3,4], grid cells [5,6] from existing continuous attractor models. It is and others.

based on oscillatory inhibitory interneural One of the most important ideas involving networks, performing rank-order the possibility of continuous spatial spatiotemporal encoding of environmental representation in brain is the hypothesis of signals. Moreover, this model exhibits a continuous neural network attractors suggested significant feature of such networks an in the range of papers [7-10]. Such a model is opportunity for multiscaled spatial able to maintain neural activity to represent any representations. Finally, the model shows the location along continuous physical manifold.

existence of three main directional grid axes in Attractor networks were shown to maintain brain, confirming corresponding conclusion both continuous and discrete patterns based on recent neurophysiologic studies on simultaneously [11,12]. Therefore such human [14,15].

networks can be used for storage of location, for instance, in continuous physical space after Methods it was determined and learnt upon the investigation of relations between objects and Oscillator ECI-networks were used as a boundaries of test environment [13].

basic neural network model. A detailed Nevertheless, the problem of functional cell description of mathematical model and network specialization and their formation in neural dynamics can be found in the papers [16-22].

dynamics of spatial behavior still remains The ECI-network architecture appears to unsolved. Moreover, modern conception be an active structure of loosely-coupled concerning neural association, proceeding nonlinear oscillators, joined together by spatial encoding and environmental signal recurrent inhibitory connections in square processing in brain, in our opinion lacks grids. In case of certain parameters and understanding of neural network dynamics as a permanent external energy inflow the network generates a wideband range of low-frequency XVI theta-rhythms, low-amplitude high-frequency found in superficial layers, in deep layers all ripples and gamma oscillations of neural units simultaneously belong to both types of membrane potential, which well corresponds to assemblies, that is they are universal modern neurophysiological experimental data (conjunctive).

[23,24]. The function of directional coding in A special feature of multilayer ECI- superficial ECI-network layers is notable for network spatial organization is the existence of high resource of implementing informational two interchangeable systems, referential and units. It means high accuracy of angular informational. Informational modules have resolution and multiscaled spatial azimuthal external variable informational inputs whereas representation. Indeed, on one hand cercal referential modules have only constant input. distribution of directional units results in higher Input impulse signals of constant angular resolution with the rise of the number amplitude and gradient signals in certain range of such units participating in distribution. On of normalized values were used as input spatial the other hand the existence of numerous signal patterns in the experiments. assemblies of directional coding and the Dynamics of transient processes and occurrence of the frequency gradient along synchronized states of neural associations of vertical network axe enable spatial coding scale different levels from oscillator quartettes to manifold. In the dippiest ECI-network layers all assemblies was assessed in the study and an units simultaneously form assemblies of both impact of temporal intervals between events on types, i.e. these layers are uniform spatial relations of different neural groups and (conjunctive).

their transition to chaotic dynamics was taken Finally, one of the main results obtained in. during model investigations is that groups (quartettes) of informational units from one Results layer have the same orientation and spacing, confirming experimental result about grid cells Computational experiments showed an that have firing fields with the same spacing assembly principle of periodic positional and and orientation in different enclosures [25].

directional encoding of input signal patterns in ECI-networks. Input signal pattern encoding is Conclusions nonlinear and depends on position of informational units in ECI-network circuit 1. On the example of informational units, determining directional and topologic implementing topological and directional specialization of these units. Continuous neural encoding, it was shown that they could network attractor dynamics is shown to comprise both specialized neural functional underline network phase states continuum. groups (assemblies) and polyfunctional Transitions between networks states can be assemblies of the same network exhibiting realized by self-motion signals: linear and universal nature.

angular velocity of navigator turn. Phase 2. Relevant spatial connection between manifold of established responses, formed due firing field of grid cells involved in path to network recurrent interactions and shifting integration is maintained almost constant in mechanism, driven by velocity inputs, form the enclosures of different scales.

base of path integration dynamic phenomenon 3. Temporal factor, i.e. dynamic in the network. Its important to mention the integration process, determines if informational group (assembly) way of navigator spatial unit will belong to one or another assembly.

Pages:     | 1 |   ...   | 55 | 56 || 58 |

2011 www.dissers.ru -

, .
, , , , 1-2 .