FR3130396A1 - Method for reducing psychological stress on a journey by selecting paths to environments suitable for the provision of well-being, evaluated algorithmically - Google Patents

Abstract

The invention covers a method of reducing psychological stress by exposing individuals to environments providing well-being on journeys from a point of origin to a point of destination, as well as the apparatus allowing its implementation. implemented. At least one map of well-being level values, making it possible to assess the quality of experience of the environments located at these points, is generated by an algorithm in conjunction with systems capable of automatic cognitive learning (for example a logistic regression system or even a neural network system) from data, such as digitized images and/or sounds, characteristics of at least one environment. This map of well-being values is then routed to at least one terrestrial navigation system which uses the geo-located well-being levels in order to determine the paths, from a point of origin to a point of destination, whose environments at crossing points are the most suitable for reducing the user's psychological stress.

Description

Method for reducing psychological stress on a journey by selecting paths to environments suitable for the provision of well-being, evaluated algorithmically

Technical field of the invention

The invention is at the crossroads of the following technical fields:

- Reduction of psychological stress by exposure to natural and pleasant environments (A61M21 / 02)

- Land route planning (G08G1/09)

- Methods and apparatus adapted to the processing of large quantities of data (big-data) for the automated generation of map data for the purpose of planning land journeys (G06K9/62)

Technical problem

Stress is a biological reaction to external stimuli, or to lack of satisfaction at work or in personal life (non-patent document 1). Although some studies have shown the usefulness of stress in human development, the long-term effects on the body can be detrimental. Stress can induce different types of reactions: psychological, physiological and behavioral (non-patent document 2). In the most serious cases, it can cause insomnia, fatigue, loss of appetite, bulimia, drug addiction, depression. Stress can be the source of serious physical or mental illnesses such as obesity, cardiovascular disease, certain forms of autism, attention deficit, Asperger's disease. It concerns all ages and all socio-cultural categories. Stress is an omnipresent scourge of the modern era, the reduction of which represents major economic, social and medical challenges.

In the field of psychological stress reduction, common solutions are the practice of sport, meditation and immersion in a relaxing environment. The notable disadvantage of these is the need to have to set aside dedicated time for these activities when time is scarce in a modern lifestyle. There is also the manipulation of de-stressing objects (malaxotherapy) but no scientific study has really proven the effects.

In the field of route planning, conventional terrestrial navigation systems commonly offer the fastest path, the shortest path, the path that avoids traffic, the path that avoids tolls. The disadvantage is that in all cases, there is no consideration of the impact of the environments crossed on the psychology and the stress of the user. Patent document 3 describes a method for recommending pleasant routes from a point of origin to a destination based on pleasantness levels of the environments on the routes. The major problem with this method is that the emotional evaluation of each location on the map is constructed by two inefficient methods. The first method described by patent 3 is by the manual evaluation of images of the place by one or more people via a survey, which prevents any scaling in a practical way on large quantities of places such as hundreds of thousands or millions of different environments. The other disadvantage of this technique is that if a place changes appearance, by the intervention of construction sites for example, the refreshment of the emotional information related to this place must also be re-evaluated manually by people. Also, another disadvantage of having the data evaluated by people rather than by an algorithm is the variability of the results, in fact the same person surveyed at two separate times will probably give a different rating for the same place, due to changes in mood or state of mind, while an algorithm will always give the same results for the same input values. The second described by patent 3 is by the systematic analysis of the text labels (tags) linked to the images of these environments to extract the emotion conveyed, the problem being that this text is not a reliable source for identifying the emotion conveyed by these environments, because it is not always directly related to the latter, or can in particular describe it in an ironic way. In addition, the methods described in patent 3 make use of images but not sounds, which can be the source of a significant difference in the experience lived in environments where, for example, noise pollution is high.

In the field of creating maps of levels of perceived well-being provided by environments, data is generally obtained by paper or online surveys, which in all cases require manual responses from people (such as than in patent document 3). There are also maps of points of interest, but they are not always suitable for stress reduction, and are more geared towards practical daily uses or tourism, for example an airport, a shopping center or a museum. They are also sparse or absent in certain geographic areas.

Presentation of the invention

To reduce the stress of an individual during a trip from a point of origin to a destination by taking advantage of his trips, it is possible according to various scientific studies (including that described in non-patent document 4), to alter the journey to expose this individual to environments more favorable to well-being, pleasure and relaxation, with a reduced investment of additional personal time.

In order to assess which are the most suitable routes to reduce stress, it is possible to use a map of levels of well-being of the environments located on the possible routes to the destination. The role of this map of levels of well-being is to represent, by values located on this map, the quality of the experience of well-being lived at these points (as illustrated in the ). Maps of well-being levels can be generated without having to resort to ratings of places developed by individuals (as in patent 3, paragraph 0023), through the use of an algorithm to determine the values of well-being of these environments based on images and/or sounds in a completely automated way. The map of these perceived levels of well-being could then be used by a terrestrial navigation system such as a GPS or a navigation application in order to determine the possible routes most likely to reduce the user's stress.

The present invention describes a method for generating maps of levels of well-being from images and/or sounds of geo-located environments, automatically and without human intervention (contrary to patent 3). The automatic evaluation of the level of well-being provided by the environments can be achieved by the use of pre-trained neural networks on image and sound data associated with different levels of well-being. This method, on the other hand, requires a very large amount of data.

Experiments have shown that it is possible to obtain better results in the assessment of the level of well-being of an environment with less large amounts of data by going through an intermediate step, using pre-neural networks. -trained on data of felt emotions associated with images and sounds related to environments, these emotions including positive and negative emotions, such as the appreciation of beauty (positive), quietness (positive), ugliness ( negative), and fear (negative) for images, and tranquility (positive), joy (positive), fear (negative), and annoyance linked to nuisances (negative) for sounds. An algorithm then uses these emotion values to deduce the well-being values.

The fact that the invention is automated from end to end combined with the fact that the data processing unit can be installed on several machines simultaneously, allows the processing of images and sounds of the environments in parallel, which allows the invention to be scalable in that it can dynamically adapt to very large amounts of data to be processed with reduced processing time. For example, it is possible to generate a map of well-being values on millions of geo-located points in an efficient way, to cover large geographical areas such as an entire region or an entire country in a few hours with enough computers for processing, compared to several months or years if the images and sounds of environments must be evaluated manually by people, which would then be very impractical (such as by the method described in patent 3).

The combined use of image and sound data by the invention provides the advantage of offering the end user paths that make more sense than those offered by inventions that use only image data (such as than in patent 3). For example, a journey on the outskirts of a park such as Central Park in New York can be visually relaxing, but the stress reduction experience of a pedestrian or cyclist could be greatly altered by the high noise pollution associated with it. to car traffic, to the incessant noise of horns and sirens of firefighters, ambulances and the police.

This presentation details a specific embodiment of the invention, by way of example and non-restrictive, aimed at a framework of use in the automotive industry. The environmental data was previously collected by a company specializing in the recording of photographs of environments and sounds at regular intervals on the roads using vehicles equipped with panoramic cameras and microphones, and placed in a database. provision at the inputs of data processing units (1), installed on each node of a cluster of computers to process environmental data in parallel.

This mode of use of the invention uses in its implementation an intermediate step of extracting the values of emotions felt associated with images and sounds related to the environments before obtaining the values of well-being felt, for better results and to decrease the amount of neural network training data required, analogously to that shown in the .

At the output of the data processing units (1), the maps with the well-being values (16) are stored in a centralized database accessible in writing by the computers forming the cluster.

To carry out the invention, it is first necessary to train two neural networks (7 and 11), for example of the COVNET type, in order to specialize them in the recognition of the emotions felt by means of images (7) and sounds (11 ). One can use for this purpose neural networks pre-trained respectively on the recognition of images and sounds, adapt them and re-train them in order to specialize them on data of emotions felt respectively images and sounds ( data sets whose method of constitution is described below). The output layers of the two neural networks must be configured so that each emotion represents (including for example beauty appreciation, tranquility, ugliness and fear for images, and tranquility, joy, fear and annoyance related to sound nuisances) has its own output value, so that each input data is noted by a value for each of the emotions.

The constitution of the training data of the neural networks is the only step of the realization of the invention which requires the evaluation of data by individuals, it only needs to be carried out once and then the The very exploitation of the invention is completely automatic and entirely excludes the involvement of individuals. To constitute the training datasets of the neural networks, it is necessary to gather a representative sample of images and sounds of all the typologies of terrestrial environments. Each image and sound in the sample must be evaluated by a group of people statistically representative of the target population. These images and sounds must be noted by each individual who assigns a note per emotion by convention between 0 and 1, 0 representing the absolute absence of the emotion and 1 representing the point of maximum feeling of the emotion. For each individual piece of data (image or sound) and each emotion, the scores assigned by the individuals are averaged. Then finally we normalize these averages by dividing them by the sum of the averages on all the emotions for each individual data, so that the sum of these normalized averages makes 1 for each data (image or sound).

The data from the environments is routed in packets to the data processing units (1) installed on the computers in the cluster, each packet containing the image and/or sound data for a single environment. Each image in a packet is evaluated, one by one, by the Emotional Image Analysis Neural Network (7) and yields a set of emotion values for each image in the packet (8). Similarly, each sound in a packet is evaluated, one by one, by the emotional sound analysis neural network (11) and yields a set of emotion values for each sound in the packet (12). The set of values obtained for a packet is averaged emotion by emotion, both for the images (9) and for the sounds (13). Each average value thus obtained for the image part is divided by the sum of the averages to obtain the averages normalized by emotion for the image part, the same operation is carried out for the averages of the sound part so that the sum of these normalized averages make 1 for the image part and 1 for the sound part.

For each environment, the normalized mean values obtained for positive emotions (such as appreciation of beauty and tranquility for images, and tranquility and joy for sounds) are added together to obtain values of goodness. - be felt, both for images (10) and for sounds (14).

To calculate the global well-being values for each environment, we give ourselves two weights whose sum is 1 (for example 0.8 and 0.2). If at the point under consideration there is image and sound, one of these two weights is associated with the image part and the other with the sound part, if there is only image or that of the sound at the point considered, a weight of 1 will be given to the category present and a weight of 0 therefore to the category absent. We multiply the weight of the image part to the value of well-being obtained for the neural processing of the image (10) then we multiply the weight of the sound part to the value of well-being obtained for the neural processing sound (14) then, finally, these two values are added to obtain the overall well-being value at the point considered, which is then added to the set of overall well-being values already calculated (15).

These global welfare values (15) are then associated with the coordinate points of the environments on the map (16).

The map of well-being values (16) once constituted, is routed for use, to the terrestrial navigation systems (17) integrated into the vehicles in this example, via for example a file transfer through the Internet, in order to be stored in the internal memory of said navigation system.

The navigation system (17) can finally use the data from the well-being map in order to modulate the driver's journey towards his destination by bringing there the dimension of stress reduction, according to the levels of well-being of the environments on the possible routes as well as the other more traditional route selection criteria, according to the driver's choices.

Industrial applications

The invention targets several industries. Here are some concrete examples of applications:

- In the health industry, and in particular connected health, the invention can be integrated into navigation systems in mobile phones for pedestrians or cyclists, in order to help users reduce their psychological stress and increase their well-being in their daily journeys. It could be recommended by doctors, as a complementary and risk-free method, to a treatment for the improvement of the reduction of the effects related to the stresses of patients.

- In the automotive industry, the integration of the invention into vehicle-integrated navigation systems would make it possible to direct the driver to his destination, through a series of environments favorable to psychological well-being, including a Conventional car navigation not equipped with the invention is not capable. The invention could then, for example, be part of the options for improving the comfort of the driver and passengers available when purchasing a vehicle.

- In the tourist industry, a travel agency could recommend personalized routes to customers between several points of interest selected by them on their destination. The invention would improve the psychological experience of customers in their travels as well as their overall satisfaction during their trip.

is a non-restrictive representation of the data flow in a particular mode of implementation of the invention. The image data (3 and 5) and (6)/or (4) of sounds corresponding to at least one geo-located environment of at least one map (2), and whose mathematical properties are described in [Math 1] in the appendices, are placed at the input of a data processing unit (1), for one environment at a time. This data processing unit (1) is composed of two pre-trained neural networks on images (7) and sounds (11) associated with different emotional values, and data processing steps output from these networks of neurons. Neural networks evaluate input environment data and assign each image (input of 7) and sound (input of 11) a value for each emotion considered (e.g. appreciation of beauty, tranquility, ugliness and fear for images, and tranquility, joy, fear, and annoyance linked to nuisances for sounds). The emotional values of the images (8) and/or sounds (12) of each environment are averaged (9 for the average of the images of the environment and 13 for the average of the sounds of the environment) and are used to calculate the values of well-being felt for the set of images (10) and sounds (14) for each environment. A single well-being value is calculated from these two values and added to the list of well-being values for all the environments of the map (15). These values are then placed on the map (16) at the geographical points for taking environmental data (3-6). The well-being data on the map (16) is then routed to at least one terrestrial navigation system (17) which can use it to determine the paths providing the most well-being from a point of origin towards a destination.

illustrates the meaning of values of well-being felt in the face of environments. The values here are arbitrarily distributed over a range of 0 to 1 (101), with 0 representing the most uncomfortable level of well-being (malaise), and 1 representing the highest level of well-being. The images (102-107) present different examples of images of environments as well as the welfare values associated therewith by the data processing unit (1). Images representing dirty (102), degraded (103), unreassuring (104) places have low well-being values, whereas a place of entertainment such as a funfair (105), a body of water at next to a building with beautiful architecture (106) or a beautiful road in a forest environment (107) have strong well-being values.

Claims (9)

Method for generating at least one map of levels of well-being linked to their respective geo-localized environment from at least one datum per environment of at least one type of datum (for example at least one image, therefore of the image type) and comprising at least the following steps:

1. A step of acquiring data relating to each environment, associated with the geolocation coordinates of the environment, from a static source (such as a database or a DVD) or dynamic (such as a stream from a camera or a microphone), which includes obtaining this information from a data acquisition module of the environments
2. A step of identification/extraction of well-being data corresponding to the characteristics of each environment
3. A step of associating well-being values with the coordinates of each environment
4. A step of generating at least one map comprising all the well-being values

Method for generating a map of well-being values according to Claim 1, characterized in that the step of identifying/extracting data from the environment is carried out by a system capable of automatic cognitive learning (for example a logistic regression system or even a neural network system) pre-trained to determine levels of well-being by type of data (for example if there are images and sounds, there will be a system capable of pre-trained automatic cognitive learning for images and one for sounds), as well as a sub-step of aggregation of these values to obtain an overall well-being value by environment Method for generating a map of well-being values according to claim 1, characterized in that in the identification/data extraction step comprises at least the following sub-steps:

1. A sub-step of analyzing and generating emotional values of environmental data through as many systems capable of pre-trained machine cognitive learning as types of data (e.g. if there are images and sounds, there will be one system capable of pre-trained cognitive machine learning for images and one for sounds), the systems capable of machine cognitive learning being trained on data conveying a set of emotions containing at least a positive emotion and a negative emotion by type of data (for example the appreciation of beauty, tranquility, ugliness and fear for images, and the tranquility, joy, fear and annoyance linked to nuisances for sounds)
2. A sub-step of calculating environmental well-being values from the emotion values obtained for the environmental data

Method for generating at least one well-being level map according to claims 1-3 characterized in that the systems capable of automatic cognitive learning used are neural network systems Method for generating at least one map of well-being levels according to claims 1-4, characterized in that the environmental data used are of the image and/or sound type. Method for generating at least one well-being level map according to claims 1-4 characterized in that the environmental data used are of the image type Method for generating at least one map of well-being levels according to claims 1-4 characterized in that the environmental data used are of sound type Method for planning at least one terrestrial journey by using a map of well-being values, characterized in that it comprises at least the following steps:

1. A step of generating a map of well-being levels according to claims 1-7
2. A step of routing the map data (for example by transfer via a network) to at least one terrestrial navigation system (such as a GPS or a navigation application)
3. A step of using the well-being levels map data via the terrestrial navigation system to determine the possible paths between departure and destination selected from the paths having their highest felt well-being total value ratings , corresponding to the paths with the most significant stress-reducing beneficial effects.

Apparatus provided with a calculation unit, memory and interfaces characterized in that it is suitable for implementing the method according to claims 1-8