Method for determining a mode of transport for a journey travelled by at least one user

Abstract

The present invention relates to a method for determining the mode of transport (MTT) of a journey. For this project, time-stamped geolocation (GPS) data of the journey is acquired, and a series of steps (SET) is implemented to determine the mode of transport (MTT). Each step in the series of steps (SET) is linked to a different category of the mode of transport, with some steps implementing models trained on time-stamped geolocation data (TSD). Figure 1 to be published

Description

Title of the invention: Method for determining a mode of transport for a journey taken by at least one user Technical field

[0001] The present invention relates to the field of determining a mode of transport of a journey traveled by a user.

[0002] The transport sector is an essential aspect of daily life on which the global economy depends heavily. However, it also plays a major role in the deterioration of air quality and the increase in CO2 emissions, particularly in urban areas. According to the World Health Organization (WHO), transport is one of the main sources of air pollution, and it is directly linked to multiple respiratory and cardiovascular diseases. In addition, global transport is responsible for approximately 24% of direct CO2 emissions from fuel combustion, of which almost three-quarters correspond to cars, trucks, buses and motorcycles, further exacerbating the phenomenon of global warming. Other types of harmful emissions include nitrogen oxides (NOx), ground-level ozone (O3) and particulate matter (PM), which generally exceed the recommended limits.The importance of transport's role in air quality is also reflected in the growing number of policy actions taken in Europe in recent years, making it a central issue. Strategies to address this issue include: vehicle efficiency and low-carbon fuel standards, development and incentives for low-emission vehicles, fiscal policies and tax regimes, and development of road and public transport infrastructure.

[0003] One of the fundamental characteristics of a trip is the mode of transport used. Indeed, modern mobility is no longer limited to walking or running, and different types of motorized and non-motorized vehicles are now available to the population. The different characteristics of the means of transport chosen reflect essential aspects of mobility, including: lifestyle, optimization of time and budget, choice of routes, pollutant emissions, etc.

[0004] This choice of users with regard to the different possible modes of transport and the distribution of these choices of the different users in a given geographical area are today poorly understood, although they are essential for the development of road transport networks, public transport or soft mobility (bicycles and scooters in particular). To date, public authorities have analysis tools to measure mobility choices and the potential impact of development work (road, public transport, soft mobility) essentially based on mobility modeling approaches. However, these models are very difficult to set up (need for very expensive population survey data, need for a significant calibration effort), to maintain (maintenance is delegated to design offices, it is therefore difficult to develop the tool quickly), to update (population surveys are conducted every 5 to 10 years).

[0005] There is therefore a need to characterize rapid changes in mobility by determining users' modes of transport based on their journeys, based on real mobility data. Prior art

[0006] Massive mobility data collection and analysis have taken a leap forward with the development of highly accessible wearable sensors, now included in devices such as mobile phones and smartwatches. Indeed, geospatial technological innovations have overcome the usual obstacles of traditional statistical surveys or interviews, for example: updates separated by years, extremely high costs, low spatio-temporal resolution and low reliability due to the human factor. Mobility data can be retrieved by different types of sensors, including: cellular network, WiFi, Bluetooth and GPS sensor, which were first made available on a mobile phone around the year 1999.Two important and very different types of mobility data are Call Detail Records (CDRs) and geolocation data (e.g., GPS, Global Positioning System); notably, the former tends to have a higher penetration rate while the latter has a much higher spatiotemporal resolution.

[0007] Applications concerning mobility data are innumerable and cover distinct fields such as: exploration of mobility patterns, urban and road planning, traffic forecasting, health and transport monitoring, etc. In this context, an essential characteristic is the transport mode used, and asking the user to provide information is problematic (lack of motivation or attention, and / or difficulties related to multimodal travel, etc.). Therefore, automatic recognition of the transport mode from GPS data has been a topic of utmost importance in transport science for several years; early research was based on a Geographic Information System (GIS) and simple heuristic rules. Such heuristic rules are described in particular in the document: STOPHER, P., CLIFFORD, E„ ZHANG, J. and FITZGERALD, C., Deducing Mode and Purpose from GPS Data, 2007, Working paper ITLS-WP-08-06. Although even more difficult, recent research also focuses on identifying the mode of transport from sparse data such as CDR data. Such a method is described in particular in the patent application whose filing number is FR 21 / 13219.

[0008] The literature on transport mode identification from GPS data is vast and constantly growing. It closely follows innovations in GPS sensors, growth in penetration rate and improvements in computing power for more efficient analyses. Given the apparent simplicity of the problem, it may seem surprising that there is no completely satisfactory solution. Indeed, the wide variation in existing techniques is evidence of the difficulty of this task. However, it is noted that most current studies are based on supervised learning methods. It can be noted that several works in the literature include multimodal journeys, which requires a division of the trajectories. The automation of this preliminary step is a research topic in its own right.

[0009] Common transportation mode detection algorithms include: artificial neural networks, Bayesian networks, convolutional neural networks, discrete hidden Markov models, decision trees, expectation maximization techniques, fuzzy logic rules, GIS (Geography Information System) based algorithms, graphical convolutional networks, K-nearest neighbors, logistic regressions, random forests, recurrent neural networks, support vector machines, unsupervised deep learning, etc.

[0010] Input variables for GPS sensors are very diverse, including for example: latitude and longitude pairs, travel time / length, functions of velocity, acceleration, jerk and / or heading signals (mean, median, standard deviation, percentiles, covariances, etc.), GPS signal quality, stopping rate, "hotspot" information, and others. Several studies have also used external GIS data including underlying layers, road networks and / or public transport data.

[0011] Finally, the output is the predicted transport mode or, sometimes, the transport mode category. Most studies are able to detect between four and five modes, and tend to include walking, car, and bus (rarely motorcycle). The accuracy can reach up to 90%, but interpreting and / or comparing the results between different studies is often meaningless; the datasets used range from a single user and a few trips to hundreds of users, the quality differs and is often unrealistic, the GPS frequency varies considerably (from 1 to 1 / 30 Hz), etc.

[0012] The fundamental works on the detection of transport modes presented below are classified in chronological order in order to reflect the natural evolution of the different techniques applied.

[0013] The first researches apply a supervised learning approach which consists of a segmentation method, an inference model (decision tree, Bayesian network,...), and a post-processing algorithm based on conditional probability. The modes considered are walking, bus, bicycle and car. A decision tree can then be proposed followed by a first-order hidden Markov model to distinguish between stationary, walking, running, bicycle and motorized modes.

[0014] For example, the document: "GONZALEZ, PA, et al., Automating mode detection for travel behavior analysis by using global positioning Systems enabled mobile phones and neural networks, 2010, Intelligent Transport Systems" suggests the use of neural networks to classify the basic modes of walking, bus and car.

[0015] Most of the developments do not allow for robust distinction between motorcycle travel and motor vehicle travel. In addition, some of the developments cannot be used in real time, because the time required and the computer memory required are significant.

[0016] Patent application FR3112398 (US 2022 / 010220) allows a method of determining a mode of transport of a journey to be determined, by a study of similarity between the acquired journey and previous journeys by means of a modified discrete Fréchet distance. The precision and robustness of the determination of the mode of transport obtained by this method can be further improved. Summary of the invention

[0017] The present invention aims to determine the mode of transport of a journey traveled by a user in real time (just at the end of the journey), in a robust manner, and taking into account a large number of possible modes of transport, including motorcycles. For this purpose, the present invention relates to a method for determining the mode of transport of a journey traveled. For this project, time-stamped geolocation data of the journey is acquired and a succession of steps is implemented until the mode of transport is determined. Each step in the succession of steps is linked to a different category of the mode of transport, with certain steps implementing Models trained on time-stamped geolocation training data. Using one step per transport mode category within a series of steps allows for greater robustness. In addition, using data-driven models allows for rapid determination, and therefore, real-time usage.

[0018] The invention relates to a method for determining a mode of transport of a journey traveled by at least one user, by means of a geolocation sensor. For this method, the following steps are implemented: a. Time-stamped geolocation data of said route traveled by said user is acquired using said geolocation sensor; b. A mode of transport of said journey is determined by means of said acquired time-stamped geolocation data, by implementing a succession of steps until said mode of transport is determined, the steps of said succession of steps being chosen from at least the following steps: i. A model trained from training time-stamped geolocation data and / or a heuristic filter constructed from training geolocation data is applied to said acquired time-stamped geolocation data to determine whether the mode of transport is motorized or not; ii. A model trained from training time-stamped geolocation data is applied to said acquired time-stamped geolocation data to determine whether the mode of transport is a motorcycle; iii. At least one mapping algorithm is applied to said acquired time-stamped geolocation data to determine whether the mode of transport is public transport not running on a road network; and iv. At least one model trained from training time-stamped geolocation data is applied to said acquired time-stamped geolocation data to determine whether the mode of transport is a vehicle traveling on a road network.

[0019] According to one embodiment of the invention, historical data of journeys of said user are acquired, and said succession of steps further comprises a step of comparing said acquired time-stamped geolocation data with said historical data, and the mode of transport is determined if the acquired time-stamped geolocation data are similar with said historical data of at least one journey, preferably a similarity is determined by means of calculating a distance, such as the Fréchet distance or the Hausdorff distance.

[0020] Advantageously, said mode of transport is determined from walking, running, bicycle, motorcycle, tram, airplane, train, subway, motor vehicle and bus.

[0021] According to one implementation, the method comprises a prior step of determining at least one descriptor of said journey from said acquired time-stamped geolocation data, and at least one step of said succession of steps takes into account said at least one descriptor.

[0022] According to one aspect, the determination of a mode of public transport not circulating on a road network is implemented by means of a mapping-pondence algorithm and the Hausdorff distance.

[0023] According to one embodiment option, data from vehicles belonging to said user are acquired, and said succession of steps is adapted as a function of said data from vehicles belonging to said user.

[0024] According to one embodiment, said models are trained from time-stamped geolocation data by machine learning using at least one neural network.

[0025] According to one implementation, said succession of steps first implements the determination of a motorized or non-motorized mode of transport.

[0026] According to one aspect, said succession of steps implements the determination of a mode of transport by motorcycle before the determination of a mode of transport by public transport not circulating on a road network and before the determination of a mode of transport circulating on a road network.

[0027] According to one option, said succession of steps implements the determination of a mode of public transport not circulating on a road network before the determination of a mode of transport circulating on a road network.

[0028] According to one embodiment, said heuristic filter is based on a statistical analysis of said time-stamped learning geolocation data.

[0029] Advantageously, a mapping algorithm is applied per mode of public transport not running on a road network.

[0030] According to one embodiment, if the mode of transport is determined to be non-motorized, said trained model and / or said heuristic filter further determines the type of non-motorized transport.

[0031] Other characteristics and advantages of the method according to the invention will appear on reading the following description of non-limiting examples of embodiments, with reference to the appended figures described below. List of figures

[0032] [Fig.l]

[0033] [Fig.l] illustrates the steps of the method according to a first embodiment of the invention.

[0034] [Fig.2]

[0035] [Fig.2] illustrates the steps of the method according to a second embodiment of the invention.

[0036] [Fig.3]

[0037] [Fig. 3] illustrates the steps of the method according to a third embodiment of the invention.

[0038] [Fig.4]

[0039] [Fig.4] illustrates the steps of the method according to a fourth embodiment of the invention.

[0040] [Fig.5]

[0041] [Fig.5] illustrates a first example of a path on which the method is applied according to an embodiment of the invention.

[0042] [Fig.6]

[0043] [Fig.6] illustrates a second example of a path on which the method is applied according to an embodiment of the invention.

[0044] [Fig.7]

[0045] [Fig.7] illustrates a third example of a path on which the method according to an embodiment of the invention is applied. Description of the embodiments

[0046] The invention relates to a method for determining a mode of transport for a journey taken by a user, by means of a geolocation sensor. In other words, for a journey taken by a user, the method determines which means of transport was used for this journey taken, for example by bicycle, bus, train, motorcycle, etc.

[0047] For the method according to the invention, the journey traveled by the user may be a monomodal journey, that is to say a journey using a single mode of transport between the departure and the arrival.

[0048] A geolocation sensor is a sensor that measures the user's position and the time of measurement. This is called time-stamped geolocation. Such a geolocation sensor may use GPS technology, Galliléo technology, or any equivalent technology. Advantageously, the geolocation sensor may be provided in the user's smartphone, or in a connected watch, or any connected object or any similar object.

[0049] The method according to the invention also implements a set of previous data of journeys traveled (by the same user and / or by other users). For this set of previous data, the time-stamped geolocation data and the mode of transport that was used are known. This set of data previous is called, in the remainder of the application, time-stamped learning geolocation data. Thus, the method according to the invention may include a step of acquiring or even measuring time-stamped learning geolocation data.

[0050] The method according to the invention comprises the following steps:

[0051] 1) Acquisition of time-stamped geolocation data

[0052] 2) Implementation of a succession of steps for determining the mode of transportation

[0053] 2.1) Determination of a non-motorized mode of transport

[0054] 2.2) Determination of a mode of transport by motorcycle

[0055] 2.3) Determination of a mode of public transport outside the road network

[0056] 2.4) Determination of a mode of transport by motorized vehicle on the road network

[0057] The steps of the succession of steps are carried out until a mode of transport is determined. In other words, once a step has identified a mode of transport, the other steps are not carried out. The steps of this succession of steps can be implemented in this order or in any different order. Step 1 is implemented by means of the geolocation sensor, and step 2 can be implemented by computer means, such as a computer, a server, or a mobile phone. These steps will be detailed in the remainder of the description. These steps can preferably be carried out in real time, that is to say immediately after the end of the journey. Indeed, the method according to the invention can be carried out in a few seconds.

[0058] [Fig. 1] illustrates, schematically and in a non-limiting manner, the steps of the method according to a first embodiment of the invention. First, GPS time-stamped geolocation data are acquired. Then, a succession of SET steps is applied to determine the mode of transport MDT. Within the succession of SET steps, at least one step is implemented until a mode of transport is determined. [Fig. 1] illustrates a possible ordering of the steps. However, other orderings are possible. Within the succession of SET steps, a step of determining a non-motorized mode of transport NMO can be implemented first. This NMO step also uses DAP learning geolocation data. If this NMO step determines a non-motorized mode of transport: "yes", then the succession of steps stops, and the mode of transport MDT is a non-motorized mode of transport.On the contrary, if this NMO step does not determine a non-motorized mode of transport: "no", then the succession of SET steps continues towards a step of determining a mode of transport by motorcycle MOT, which also uses DAP learning geolocation data. If this MOT step determines a mode of transport by motorcycle: "yes", then the succession of SET steps stops, and the mode of transport . MDT is a mode of transport by motorcycle. On the contrary, if this step MOT does not determine a mode of transport by motorcycle: "no", then the succession of steps SET continues towards a step of determining a mode of public transport outside the road network TRC (for example: train, plane, metro, tram). If this step TRC determines a mode of public transport outside the road network: "yes", then the succession of steps SET stops, and the mode of transport MDT is a mode of public transport outside the road network. On the contrary, if this step TRC does not determine a mode of public transport outside the road network: "no", then the succession of steps SET continues towards a step of determining a mode of transport by motorized vehicle on the road network VER (for example: bus or car), which also uses DAP learning geolocation data.If this step determines a mode of transport by motorized vehicle on the road network: “yes”, then the succession of SET steps stops and the mode of transport MDT is a mode of transport by motorized vehicle on the road network.

[0059] According to one embodiment of the invention, the method may comprise a prior step of determining at least one descriptor of said journey traveled from the acquired time-stamped geolocation data. Non-limiting examples of such a descriptor may be the user's speed, the user's acceleration, the specific power, etc. For this embodiment, the steps of the succession of steps may take into account at least one descriptor of the determined journey. Indeed, such a descriptor promotes the implementation of the different steps of the succession of steps.

[0060] For this embodiment, the method may comprise the following steps:

[0061] 1) Acquisition of time-stamped geolocation data

[0062] 1') Determination of at least one descriptor

[0063] 2) Implementation of a succession of steps for determining the mode of transportation

[0064] 2.1) Determination of a non-motorized mode of transport

[0065] 2.2) Determination of a mode of transport by motorcycle

[0066] 2.3) Determination of a mode of public transport outside the road network

[0067] 2.4) Determination of a mode of transport by motorized vehicle on the road network

[0068] Step 1' may also be implemented by computer means, in particular a computer, a server or a mobile phone. These steps will be detailed in the remainder of the description.

[0069] Advantageously, the step of the succession of steps for determining a mode of public transport outside the road network can be carried out by means of at least one geographic information system (GIS). Such a geographic information system can include the routes of at least one non-road public transport mode, e.g. railways, metro lines, airport location, etc.

[0070] [Fig.2] illustrates, schematically and in a non-limiting manner, the steps of the method according to this second embodiment. This figure illustrates a combination of the two configurations described above (GIS, and prior step of determining a descriptor), however, these embodiments are independent and can be applied alone. The steps described in relation to [Fig.l] are not detailed again. For this second embodiment, the method comprises an additional step of determining at least one DES descriptor of the route traveled from the GPS time-stamped geolocation data. This step of determining at least one DES descriptor is also applied to the DAP learning geolocation data. The at least one DES descriptor is implemented in the different steps of the succession of SET steps.In addition, a GIS geographic information system is used for the step of determining a mode of public transport outside the TRC road network.

[0071] According to an implementation of the invention, the method can implement a step of acquiring data from the user's vehicles, and can adapt the sequence of steps by adapting the steps according to the data from the user's vehicles. For example, it is possible to acquire, in particular by means of a human-machine interface, all the vehicles belonging to a user, for example: motorcycle, bicycle, motor vehicle, etc. This data can then be used in the sequence of steps in order to improve the accuracy of the determination of the mode of transport. Indeed, a mode of transport for which the user owns a vehicle becomes more likely.

[0072] According to one aspect of the invention, the method may implement a step of acquiring the user's journey history. The user's journey history may include time-stamped geolocation data with which a mode of transport is associated. In this case, the succession of steps may include an additional step of determining the mode of transport by comparing the journey traveled with the user's historical journey data. Thus, the accuracy of determining the mode of transport may be improved, in particular for the user's regular journeys: for example, home-work journeys. This additional step may also make it possible to reduce the time taken to determine the mode of transport, since fewer steps in the succession of steps may be useful.

[0073] According to this aspect of the invention, the method may comprise the following steps:

[0074] 1) Acquisition of time-stamped geolocation data

[0075] 2) Implementation of a succession of steps for determining the mode of transportation

[0076] 2.1) Determination of a non-motorized mode of transport

[0077] 2.2) Determination of a mode of transport by motorcycle

[0078] 2.3) Determination of a mode of public transport outside the road network

[0079] 2.4) Determination of a mode of transport by motorized vehicle on the road network

[0080] 2.5) Comparison with a history

[0081] Step 2.5 may also be implemented by computer means, in particular a computer, a server or a mobile phone. These steps will be detailed in the remainder of the description.

[0082] [Fig. 3] illustrates, schematically and in a non-limiting manner, the steps of the method according to this third embodiment. The steps described in relation to [Fig. 1] or [Fig. 2] are not detailed again. This figure illustrates a combination of the two configurations described above (acquisition of the user's vehicles and acquisition of the user's journey histories), however, these embodiments are independent and can be applied alone. For this embodiment, at least one vehicle belonging to the user GAR is acquired, and in at least one step of the succession of steps, for example the step of determining a mode of transport by motorcycle MOT, the vehicle belonging to the user is taken into account.In addition, at least historical data of the user's journeys HIS with associated transport mode are acquired, and the succession of steps SET comprises an additional step of COM comparison of the acquired GPS geolocation data and the historical HIS data. For the illustrated embodiment, this step of COM comparison is carried out between the step of determining a mode of transport by motorcycle MOT and the step of determining a mode of public transport outside the road network TRC. However, this step can be carried out at another location in the succession of steps, for example first. If this COM comparison makes it possible to identify a "yes" mode of transport, then the mode of transport MDT is the one identified by the COM comparison, and the succession of steps SET ends.On the contrary, if this COM comparison does not make it possible to identify a “non” mode of transport, then the succession of SET steps continues with the following step, for the embodiment illustrated by the step of determining public transport outside the TRC road network.

[0083] According to one embodiment of the invention, the step of determining a mode of public transport outside the road network may comprise several sub-steps, each sub-step relating to the determination of a single mode of public transport outside the road network. In other words, the step of determining a mode of public transport outside the road network may comprise a sub-step of determining a mode of transport by plane, and / or a sub-step of determining a mode of transport by train and / or a sub-step of determining a mode of tram transport and / or a sub-step of determining metro transport, etc.

[0084] According to an implementation of the invention, the step of determining a mode of motorized transport on the road network may comprise several sub-steps, each sub-step relating to the determination of a single mode of motorized transport on the road network. In other words, the step of determining a mode of motorized transport on the road network may comprise a sub-step of determining a mode of transport by motor vehicle and / or a sub-step of determining a mode of transport by bus.

[0085] [Fig. 4] illustrates, schematically and in a non-limiting manner, the steps of the method according to this fourth embodiment. The steps described in relation to Figures 1 to 3 are not detailed again. This figure illustrates a combination of the two configurations described above (sub-steps for determining the mode of public transport and sub-steps for determining the mode of motorized transport on the road network), however, these embodiments are independent and can be applied alone. After the step of determining a mode of transport by motorcycle MOT, the succession of steps SET includes a step of determining a mode of public transport outside the road network TRC.This step of determining a mode of public transport outside the road network TRC comprises a sub-step of determining a mode of transport by plane AVI, a sub-step of determining a mode of transport by tram TRM and a sub-step of determining a mode of transport by train TRN. Each of these sub-steps can take into account data from a geographic information system GIS. In the illustrated embodiment, each sub-step can implement a different geographic information system, denoted respectively SIG1, SIG2, SIG3. Alternatively (not shown), a single geographic information system GIS can be used for the three sub-steps. If the sub-step AVI determines a mode of public transport by plane: "yes", then the succession of steps SET stops, and the mode of transport MDT is a mode of transport by plane.On the contrary, if this sub-step AVI does not determine a public transport mode by plane, then the succession of steps SET continues towards a sub-step for determining a transport mode by tram TRM. If this sub-step TRM determines a public transport mode by tram: "yes", then the succession of steps SET stops, and the transport mode MDT is a transport mode by tram. On the contrary, if this sub-step TRM does not determine a public transport mode by tram, then the succession of steps SET continues towards a sub-step for determining a transport mode by train TRN. If this sub-step TRN determines a public transport mode by train: "yes", then the succession of steps SET stops and the transport mode MDT is a transport mode by train. On the contrary, if this sub-step TRN does not determine . not a mode of public transport by train, then the sequence of steps SET continues to a step of determining a mode of transport by vehicle on a road network VER. This step of determining a mode of transport by vehicle on a road network VER includes a sub-step of determining a mode of transport by motor vehicle AUT, and a sub-step of determining a mode of transport by bus BUS. If this sub-step AUT determines a mode of transport by motor vehicle: "yes", then the sequence of steps SET stops, and the mode of transport MDT is a mode of transport by motor vehicle. On the contrary, if this sub-step AUT does not determine a mode of transport by motor vehicle, then the sequence of steps SET continues to a sub-step of determining a mode of transport by bus BUS.If this BUS sub-step determines a bus public transport mode: "yes", then the succession of SET steps stops, and the MDT transport mode is a bus transport mode.

[0086] The sequence of substeps in steps TRC and VER are illustrated in one order, however the substeps may be performed in a different order.

[0087] Furthermore, the embodiment of [Fig.4] can be combined with one of the configurations of [Fig.3]: acquisition of data from vehicles belonging to the user and / or acquisition of historical data of journeys traveled by the user and implementation of a comparison step.

[0088] 1) Acquisition of time-stamped geolocation data

[0089] During this step, time-stamped geolocation data of the path traveled by the user are acquired using the geolocation sensor. The acquisition of the time-stamped geolocation data may comprise implementing the measurement during the path traveled of the time-stamped geolocation. In this case, during this step, time-stamped geolocation data of the path traveled by the user are measured using the geolocation sensor.

[0090] According to one embodiment of the invention, the time-stamped geolocation data of the journey taken by the user can be recorded on a smartphone, a connected object, a computer or a server.

[0091] According to one aspect of the invention, this step of acquiring the time-stamped geolocation data may comprise a step of pre-processing the data. The pre-processing of the data may comprise the removal of aberrant data, the completion of the trajectory, for example by an interpolation method, and the filtering of the data. Thus, the geolocation data are more easily exploitable for the following steps, which allows for better robustness and precision in determining the mode of transport. 1') Determination of at least one descriptor

[0092] During this optional step, at least one descriptor of the path is determined. using time-stamped geolocation data acquired from the journey traveled (and possibly pre-processed). A descriptor is a characteristic of the journey that is deduced from the time-stamped geolocation data. This optional step makes it possible to obtain characteristics of the journey traveled, which can help determine the mode of transport. For example, an average speed can be a useful characteristic for determining a non-motorized mode of transport. A descriptor can be defined by a function as a function of time. This could be acceleration as a function of time.

[0093] According to one aspect of the invention, the method may further determine at least one predescriptor. The predescriptors are optionally multidimensional vectors that are used to define the one-dimensional scalars. Important functions of the predescriptors may include, among others: distances between consecutive points obtained by the geolocation sensor, total cumulative distance, speed, acceleration, specific power and jolt signals, stop and near-stop positions (a short partial stop of a journey, which may correspond to a pause induced by a red light), etc.

[0094] Advantageously, the important functions of the descriptors may be the following: functions of speed, acceleration, specific power and jerk, such as means, standard deviations, maxima, means over positive / negative points or after stopping, different percentiles, entropies; stopping frequency, mean / deviation of the distance between stopping points; several different measures of the quality of the geolocation signal, etc. Therefore, this step gives rise to a large number of preprocessed values ​​that describe the main properties of the traveled path.

[0095] 2) Implementation of a succession of steps for determining the mode of transportation

[0096] During this step, a mode of transport for the journey traveled is determined using the time-stamped geolocation data, and possibly using the descriptors determined in step 1', by means of a succession of steps. The steps in the succession of steps are carried out until a mode of transport is determined. In other words, once a step has identified a mode of transport, the other steps are not carried out. Each step makes it possible to determine a category of mode of transport. The specificity of each step allows for greater precision in the determination. This succession of steps can be considered as a "divide and conquer" sequence. This succession of steps makes it possible not to have to carry out all the steps for each journey, which simplifies the implementation of the determination of the mode of transport.Moreover, at each stage, the number of possible modes of transport is increasing. restricted which promotes reliability and robustness in determining the mode of transport.

[0097] The steps of the succession of steps can be carried out in any order.

[0098] Some of the steps in the succession of steps may implement time-stamped learning geolocation data. Preferably, a step of determining at least one descriptor may be implemented for the time-stamped learning geolocation data, preferably in accordance with step 1') described above. For this implementation, the term learning geolocation database includes the at least one determined descriptor.

[0099] Advantageously, the succession of steps can first implement the determination of the motorized or non-motorized mode of transport. Thus, a first discrimination of the mode of transport can be implemented. And after this step, if the journey traveled is not determined in non-motorized mode of transport, only motorized modes of transport remain.

[0100] Advantageously, the succession of steps can implement the determination of a mode of transport by motorcycle before the determination of a mode of common transport outside a road network, and before the determination of a mode of transport by motorized vehicle on a road network. Thus, this makes it possible to discriminate a trip by motorcycle before the other modes of motorized transport, which promotes the detection of this mode of transport. Alternatively, this step can be implemented after the step of determining a mode of public transport outside the road network.

[0101] According to one aspect, the succession of steps may implement the determination of a mode of public transport outside the road network before the step of determining a mode of motorized transport on the road network. Thus, for the step of determining a mode of motorized transport on the road network, there should only remain journeys using roads and not dedicated lanes. Alternatively, the determination of a mode of public transport outside the road network may be implemented after the step of determining a mode of motorized transport on the road network.

[0102] For the embodiment for which a history of the user's journeys with associated mode of transport is known, the step of comparing the historical data and the acquired time-stamped geolocation data can be implemented at the beginning of the succession of steps, to avoid carrying out other steps, if the comparison step makes it possible to directly determine a mode of transport. Alternatively, this step can be carried out directly after the step of determining the modes of motorized transport and / or directly after the step of determining a mode of transport by motorcycle. The earlier this step is implemented in the succession of steps, the more it can avoid carrying out a number of steps if this comparison determines a mode of transport.

[0103] The modes of transport determined by the method according to the invention can be chosen from:

[0104] - walking, running, cycling, scootering by the step of determining introduction of a non-motorized mode of transport,

[0105] - the motorcycle by the step of the mode of determining transport by motorcycle,

[0106] - the tram, the plane, the train (including the Hyperloop type train which can be (translated as "vacuum tube train"), the metro by the step of determining a mode of public transport outside the road network (such modes of public transport travel on dedicated tracks different from roads), and

[0107] - the motor vehicle, the bus by the step of determining a transport in motorized vehicle on the road network.

[0108] 2.1) Determination of a non-motorized mode of transport

[0109] During this step, it is determined, based on the geolocation data acquired in step 1, whether the journey traveled is made in a non-motorized mode of transport, by means of a model trained from the time-stamped learning geolocation data and / or by means of a heuristic filter constructed from the time-stamped learning geolocation data.

[0110] According to one embodiment, the trained model and / or the heuristic filter can further determine a probability of a non-motorized mode of transport, chosen in particular between walking, running and cycling. A mode of transport can then be determined in the case of a probability greater than a predetermined threshold.

[0111] According to one implementation, this step may implement both the trained model and the heuristic filter for verification of the results, which allows good reliability of the detection of a non-motorized or motorized mode of transport.

[0112] A trained model is a model built from data. Preferably, the model may be trained from time-stamped geolocation data from machine learning. In other words, the time-stamped geolocation data from training forms a learning base for machine learning. Preferably, the machine learning may be an artificial neural network. Alternatively, the machine learning may implement other machine learning methods.

[0113] For this automatic learning, the following steps can be implemented:

[0114] 1) initial step of selecting data in the learning base to select select data from previous journeys used for machine learning (related to discrimination between motorized and non-motorized transport modes) motorized) and, if the amount of data is insufficient, data weighting methods are also considered,

[0115] 2) selection of descriptors using mutual information statistics; in generally, between 6 and 20 descriptors can be selected, for example the descriptors can be chosen from the distances between consecutive points obtained by the geolocation sensor, total cumulative distance, speed, acceleration, specific power and shaking signals, stop and near-stop positions, etc.,

[0116] 3) standard artificial neural network architecture defined by a layer input with a node for each descriptor, at least one hidden layer, and an output layer capable of providing a probability of each non-motorized mode of transport (in a non-limiting example a probability of walking, a probability of running and a probability of cycling).

[0117] According to a non-limiting example, the neural network may comprise an input layer with a number of nodes corresponding to the number of descriptors, two hidden layers of size eight and four respectively, and an output layer providing the probabilities of each mode. The activation functions may be the rectified linear unit (ReLU) and Softmax, the optimizer may be the adaptive moment estimation (Adam), the loss function may be the categorical cross entropy, the maximized metric precision, the L2 regularization to avoid overfitting.

[0118] The heuristic filter is based on a statistical analysis of said training geolocation data. Thus, the heuristic filter is a set of simple and conservative conditions on selected features and based on statistical analyses. Its main objective is to avoid outlier errors that can disrupt the user experience. The output is a vector of the possible modes remaining after elimination.

[0119] For the embodiment implementing a heuristic filter, only the transport modes identified as possible by this heuristic filter can be considered for the rest of the succession of steps. Thus, the number of steps implemented in the method can be limited.

[0120] 2.2) Determination of a mode of transport by motorcycle

[0121] During this step, we determine, based on the geolocation data acquired in step 1, if the journey is made in a motorcycle mode of transport, using a model trained from the time-stamped geolocation training data.

[0122] A trained model is a model built from data. This model can provide a probability of a journey traveled by motorcycle. Preferably, the model can be trained from time-stamped geolocation data from machine learning. In other words, The time-stamped geolocation training data forms a training base for machine learning. Preferably, the machine learning can be an artificial neural network.

[0123] For this automatic learning, the following steps can be implemented:

[0124] 1) initial step of selecting data in the learning base to select selecting previous journey data used for machine learning (related to motorcycle transport discrimination) and, if the amount of data is insufficient, data weighting methods are also considered,

[0125] 2) selection of descriptors using mutual information statistics; in generally, between 6 and 20 descriptors can be selected, for example the descriptors can be chosen from the distances between consecutive points obtained by the geolocation sensor, total cumulative distance, speed, acceleration, specific power and jerk signals, stop and near-stop positions, etc.,

[0126] 3) standard artificial neural network architecture defined by a layer input with one node for each descriptor, at least one hidden layer, and an output layer that can provide a probability of the mode of transportation by motorcycle.

[0127] According to a non-limiting example, the artificial neural network may comprise an input layer with a number of nodes corresponding to the number of descriptors, two hidden layers of size eight and four respectively, and an output layer providing the probability of the mode of transport by motorcycle. The activation functions may be the rectified linear unit (ReLU) and Softmax, the optimizer may be the adaptive moment estimation (Adam), the loss function may be the categorical cross entropy, the maximized metric precision, the L2 regularization to avoid overfitting.

[0128] For the embodiment for which the user's vehicle data is acquired, if the user indicates that they own a motorcycle then the probability of the motorcycle mode of transportation obtained by the trained model can be increased.

[0129] If the probability determined by the trained model (possibly adjusted according to the vehicles belonging to the user) is greater than a predefined threshold, the mode of transport is determined as a motorcycle.

[0130] 2.3) Determination of a mode of public transport outside the road network

[0131] During this step, we determine, based on the geolocation data acquired in step 1, if the journey is made in a mode of public transport outside the road network, using at least one mapmatching algorithm (from the English "mapmatching" which can also be translated as geographic correspondence). For this mapmatching algorithm, a geographic information system (GIS) can be used. Here Maps™, Google Maps™, OpenStreetMap™, etc. are examples of Geographic Information System (GIS). Such a Geographic Information System (GIS) may contain dedicated lanes for off-road public transport modes, such as railways, airport locations, etc. In this step, the route traveled is compared with a route or location identified by the GIS for such a mode of transport. If the route traveled is similar to a specific route identified by the GIS, then the mode of transport is the one that corresponds to the specific route.

[0132] According to one embodiment of the invention, the step of determining a mode of public transport outside the road network may comprise several sub-steps, each sub-step relating to the determination of a single mode of public transport outside the road network. In other words, the step of determining a mode of public transport outside the road network may comprise a sub-step of determining a mode of transport by plane, and / or a sub-step of determining a mode of transport by train and / or a sub-step of determining a mode of transport by tram and / or a sub-step of determining transport by metro, etc.

[0133] According to one embodiment, a mapping algorithm can be applied per mode of public transport outside the road network. In other words, for each sub-step of this step, a mapping algorithm can be applied. For this embodiment, a geographic information system can be provided per mode of transport, or a single geographic information system which includes all the modes of transport considered.

[0134] According to one implementation, the similarity between the traveled path and the specific lane identified by the GIS can be obtained by means of a determined distance between the path and the lane identified for example by means of the Hausdorff distance.

[0135] For the possible sub-step of determining a mode of transport by plane, the following operations can be implemented. Contrary to what one might think, the detection of planes is not trivial since these journeys are generally represented by very poor geolocation data, sometimes visually indistinguishable from train journeys. The sub-step of determining a mode of transport by plane can be based on cartographic and standard conditions. The standard conditions can be based on descriptors such as the total cumulative distance and the average speed or speed above a filtered signal. The cartographic conditions can be based on intersections and can use the complete trajectory, relatively small neighborhoods around the starting and ending points of the trajectories, and the geographical position of several airports around the world.Accordingly, this substep is capable of providing a binary response indicating whether the path corresponds to an aircraft or not. In the first case, according to one embodiment, a degree of confidence may also be given (e.g. low, medium or high).

[0136] For the possible sub-step of determining a mode of transport by tram, the following operations can be implemented. Tram detection by map-matching techniques is considered complex. Indeed, taking into account railway networks can lead to considerable computational time and efforts have been made to reduce it. According to one embodiment of the invention, this sub-step may contain an additional trajectory cleaning operation, for example to remove repeated geolocation points. The main tool for map matching in this case may be the Hausdorff distance between the tram railway network and the traveled path (with the acquired time-stamped geolocation data). For this sub-step, the traveled path may be called a trajectory. The Hausdorff distance is a popular choice for measuring the similarity between geometric objects.In more detail, this sub-step can follow the following steps: 1) a dilation of the trajectory is intersected with the tram network, 2) the Hausdorff distance between the trajectory and this intersection is measured, 3) a tram overlap is defined as the ratio between: 3.1) the number of points of the trajectory inside a dilated version of the previous intersection (the dilation depends on the signal accuracy), and 3.2) the number of points of the trajectory, 4) the weighted Hausdorff distance is defined as the ratio between the Hausdorff distance and the rail coverage determined in 3). Finally, a threshold on the distance is set to determine whether a trajectory is carried out by a tram or not, in other words whether the mode of transport is tram or not.

[0137] For the possible sub-step of determining a mode of transport by train, the same operations can be implemented as those for the tramway. The difference may lie in the setting of the parameters, in particular, the distance threshold can be considerably larger.

[0138] For the possible sub-step of determining a mode of transport in the metro, in particular in the overhead metro, the same operations can be implemented as those for the tramway. The difference may lie in the setting of the parameters, for example the distance threshold.

[0139] 2.4) Determination of a mode of transport by motorized vehicle on the road network

[0140] During this step, we determine, based on the geolocation data acquired in step 1, if the journey is made in a mode of transport by motorized vehicle on the road network, using at least one model trained from the time-stamped geolocation training data.

[0141] A trained model is a model built from data. Preferably, the model can be trained from the time-stamped geolocation data of machine learning. In other words, the time-stamped geolocation data of training forms a learning base for machine learning. Preferably, Machine learning can be an artificial neural network.

[0142] For this automatic learning, the following steps can be implemented:

[0143] 1) initial step of selecting data in the learning base to select select data from previous journeys used for machine learning (related to discrimination between motorized transport modes on the road network) and, if the amount of data is insufficient, data weighting methods are also considered,

[0144] 2) selection of descriptors using mutual information statistics; in generally, between 6 and 20 descriptors can be selected, for example the descriptors can be chosen from the distances between consecutive points obtained by the geolocation sensor, total cumulative distance, speed, acceleration, specific power and jerk signals, stop and near-stop positions, etc.,

[0145] 3) standard artificial neural network architecture defined by a layer input with one node for each descriptor, at least one hidden layer, and an output layer that can provide a probability of the mode of transportation by motorcycle.

[0146] According to a non-limiting example, the artificial neural network may comprise an input layer with one node per descriptor, two hidden layers of size eight and four, and an output layer providing the probability of the mode of transport by bus or by motor vehicle. The activation functions may be the rectified linear unit (ReLU) and Softmax, the optimizer may be the adaptive moment estimation (Adam), the loss function may be the categorical cross entropy, the maximized metric precision, the L2 regularization to avoid overfitting.

[0147] According to one embodiment of the invention, this step may comprise a sub-step of determining a mode of transport by motor vehicle and a sub-step of determining a mode of transport by bus. Thus, it is easier to discriminate the mode of transport used.

[0148] For the possible sub-step of determining a mode of transport by motor vehicle, the following operations can be implemented: a product of two neural networks can be implemented: a first neural network trained to distinguish the motor vehicle from the bus (for this first neural network, the descriptors can be in particular the number of stops per kilometer, the average and standard deviation of the speed, a percentile of the highest speeds, the specific power, etc.), and the second neural network trained to distinguish the motor vehicle from the motorcycle (for this second neural network, the descriptors can be in particular the average of the acceleration for the points where the speed is positive, the average acceleration after a stop, the specific power, the maximum acceleration, etc.). Thanks to these two neural networks, the determination of a mode of transport by motor vehicle is reliable.

[0149] For the possible sub-step of determining a bus transport mode, the operations described below can be implemented. A neural network can be implemented that determines one or two specific characteristics. For the first characteristic, a score can be determined based on the proximity of the trajectory of the geolocation data with nearby bus stops (which can be obtained by cartographic information, for example by means of a geographic information system). For the second characteristic, a ratio between the number of bus journeys and the total number of journeys made by the user can be determined, if applicable (for the embodiment for which historical journey data of the user is acquired). 2.5) Comparison with a history

[0150] During this optional step, historical data of the user's journeys are acquired. The user's journey history may include, for previous journeys, the user's time-stamped geolocation data associated with a mode of transport. In this case, the sequence of steps may include an additional step of determining the mode of transport by comparing the journey traveled with the user's historical journey data. Thus, the accuracy of the determination of the mode of transport may be improved, in particular for the user's regular journeys: for example, home-work journeys. This additional step may also reduce the time taken to determine the mode of transport, since fewer steps in the sequence of steps may be useful.

[0151] For this comparison, the mode of transport can be determined by means of a similarity between the journey traveled and at least one journey in the history. This similarity can be determined by a distance between the journey traveled and a journey in the history, the distance can be for example a Hausdorff distance (in particular when comparing speed signals) or a Fréchet distance (in particular when comparing geolocation data).

[0152] According to one embodiment of the invention, certain data collection rules can be set to find a subset of candidate history paths among those made by the user. Then, these paths can be classified according to a notion of similarity and overlap between the pairs of paths. The notion used can be a modified version of the discrete Fréchet distance which has been described in particular in patent application FR3112398 (US 2022 / 010220), in particular in the part “Determination of the modified discrete Fréchet distance”. This distance makes it possible to measure the similarity between two multi-dimensional trajectories, possibly of different sizes, on the basis of an optimized association of their points. This association can be illustrated by the analogy of determining the minimum size of the leash that a man needs to walk his dog without reverse, and the distance can be calculated using dynamic programming techniques. The formula for the continuous version of the Fréchet distance between two curves A and B can be given by

[0153] F (A, B) = inf max d(A(a(t)), B( ajSre[0,l]

[0154] where "inf" denotes the infimum of a set of values, d is the Haversine distance between coordinate points, and where a and [3 are "parameterizations" of the curves A and B respectively. The calculation of the Fréchet distance between two finite sequences of points is possible via its discrete version, which can be solved by dynamic programming. Moreover, the so-called optimal matching between trajectories can also be obtained explicitly. For example, instead of considering the maximum, one can use the average distance which is more robust to the remaining aberrations in the geolocation trajectories. As a by-product, the same association of points gives rise to a notion of trajectory coverage. Finally, in order to avoid mislabeling of spatially similar trips made on different modes of transport, one can reinforce the method with the so-called Hausdorff distance between the corresponding speed signals.If this step is able to find a satisfactory historical trip candidate, the transportation mode of that historical trip is assigned to the trip traveled.

[0155] The invention thus relates to the analysis of the mobility of people, and more specifically of their choice of mode of transport for their journeys, in order to observe the impacts of mobility on pollutant emissions and air quality. It can thus make it possible to determine pollutant emissions.

[0156] Another object of the invention is to monitor the evolution of the transport mode choices of people during their travels over time so as to evaluate the strategies implemented by the public authorities (road developments, public transport, soft mobility) and to guide the choices of development of road developments, public transport and soft mobility according to the transport modes of the users. The invention can, to do this, be used to determine the transport mode choices of users in a geographical area more regularly (e.g. every month), without having to wait for new statistical surveys (household-travel surveys conducted every 5 to 10 years).

[0157] As goes without saying, the invention is not limited to the embodiments described above as examples; on the contrary, it encompasses all variant embodiments. Examples

[0158] The characteristics and advantages of the method according to the invention will appear more clearly on reading the application examples below.

[0159] For the various examples described below, the method is applied according to the embodiment of [Fig.4], with regard to the ordering of the steps of the succession of steps, by adding the comparison step to the user's historical journey data (step 2.5) as illustrated in [Fig.3]. In addition, the method implements a pre-processing step of the acquired time-stamped geolocation data. For this method, the geolocation data used for learning represent approximately 90 million km traveled with an identified mode of transport. In addition, the geographic information system is OpenStreetMap™.

[0160] In a first example, time-stamped geolocation data of a first journey is acquired using a GPS system on a smartphone. Identifying the mode of transport for this trajectory is particularly difficult: a short and relatively slow trajectory within a city and close to railway networks. For the acquired journey, a total duration of approximately 25 min and a total distance of approximately 7 km are obtained. [Fig.5] illustrates, schematically, on a road map the acquired journey between a starting point D and an arrival A. This [Fig.5] also illustrates a descriptor of the acquired journey: the vehicle speed. However, despite the pre-processing, the recorded GPS points remain affected by noise and the associated characteristics of the journey can be misleading when considered in isolation. This further complicates the task of determining the mode of transport.First, the heuristic filter works with the neural network for the non-motorized mode of transport to eliminate the non-motorized modes of transport: walking, running, cycling and scooter. Indeed, the probability of a motorized mode of transport given by this model is close to one. Then, the neural network for a motorcycle mode of transport determines a probability of 0.88, which is well above the corresponding predetermined threshold set at 0.65. Therefore, the succession of steps stops at this point and determines the mode of transport "motorcycle". The user approved this classification manually. Therefore, the method according to the invention makes it possible to identify the correct mode of transport used.

[0161] In a second example, time-stamped geolocation data of a second journey is acquired using a GPS system on a smartphone. [Fig.6] schematically illustrates on a road map the journey acquired between a starting point D and an arrival point A. This [Fig.6] also illustrates a descriptor of the acquired journey: the speed of the vehicle. First, the heuristic filter works with the neural network for the non-motorized mode of transport to eliminate the modes of transport: walking, running, cycling and scooter. Indeed, the probability of a motorized mode of transport given by this model is close to 0.65. Then, the neural network of a motorcycle mode of transport determines a probability of 0.2, which is well below the corresponding predetermined threshold fixed at 0.65. Then, the comparison step (using the discrete Fréchet distance) with historical journey data is not conclusive. Then, the mapping algorithms with plane and train respectively do not allow to determine the mode of transport: no correspondence is found between polygons of an airport or with train tracks. Then, a correspondence is found by the mapping algorithm with the tram, with a correspondence distance of 0.3 units and a rail coverage of 70%. Therefore, the succession stops at this stage and determines the mode of transport "tram". The user approved this classification manually. Therefore, the method according to the invention allows to identify the correct mode of transport used.

[0162] In a third example, time-stamped geolocation data of a third journey is acquired using a GPS system on a smartphone. [Fig.7] schematically illustrates on a road map the journey acquired between a starting point D and an arrival point A. This [Fig.7] also illustrates a descriptor of the acquired journey: the speed of the vehicle. First, the heuristic filter works with the neural network for the non-motorized mode of transport to eliminate the modes of transport: walking, running, cycling and scooter. Indeed, the probability of a motorized mode of transport given by this model is close to 0.7. Then, the neural network for a motorcycle mode of transport determines a probability of 0.3, which is well below the corresponding fixed threshold of 0.65.Then, the comparison step (using the discrete Fréchet distance) with historical journey data makes it possible to identify a similarity with a historical journey made by motor vehicle. Consequently, the succession stops at this stage and determines the mode of transport "motor vehicle". The user approved this classification manually. Consequently, the method according to the invention makes it possible to identify the correct mode of transport used. It can also be noted that if the succession of steps had continued, the same mode of transport would have been identified: the plane, tram and train connections are beyond the connection threshold, unlike the product of the two neural networks for identifying a mode of transport by motor vehicle.

[0163] These three examples further show that the method according to the invention makes it possible to identify several different modes of transport reliably.

Claims

1. Claims Method for determining a mode of transport of a journey traveled by at least one user, by means of a geolocation sensor, characterized in that the following steps are implemented: a. Time-stamped geolocation data (GPS) of said route traveled by said user is acquired by means of said geolocation sensor; b. A mode of transport (MDT) of said journey is determined by means of said acquired time-stamped geolocation data (GPS), by implementing a succession of steps (SET) until said mode of transport is determined, the steps of said succession of steps (SET) being chosen from at least the following steps: i. A model trained from training time-stamped geolocation data (TTD) and / or a heuristic filter constructed from said training geolocation data (TTD) is applied to said acquired time-stamped geolocation data (GPS) to determine whether the mode of transport is motorized or non-motorized (NMO); ii. A model trained from training time-stamped geolocation data (DAP) is applied to said acquired time-stamped geolocation data (GPS) to determine whether the mode of transport is motorcycle (MOT); iii. At least one mapping algorithm is applied to said acquired time-stamped geolocation data (GPS) to determine whether the mode of transport is public transport (TRC) not running on a road network; and iv. At least one model trained from training time-stamped geolocation data (DAP) is applied to said acquired time-stamped geolocation data (GPS) to determine whether the mode of transport is a vehicle circulating on a road network (VER).

2. Method according to claim 1, in which historical data (HIS) of journeys of said user are acquired, and said succession of steps further comprises a step of comparing (COM) said acquired time-stamped geolocation data (GPS) with said historical data, and the mode of transport is determined if the acquired time-stamped geolocation data are similar with said historical data of at least one journey, preferably a similarity is determined by means of calculating a distance, such as the Fréchet distance or the Hausdorff distance.

3. Method according to one of the preceding claims, wherein said mode of transport (MDT) is determined from among walking, running, bicycle, motorcycle, tram, airplane, train, subway, motor vehicle and bus.

4. Method according to one of the preceding claims, in which the method comprises a prior step of determining at least one descriptor (DES) of said route from said acquired time-stamped geolocation data (GPS), and at least one step of said succession of steps (SET) takes into account said at least one descriptor (DES).

5. Method according to one of the preceding claims, in which the determination of a public transport mode (TRC) not circulating on a road network is implemented by means of a cartospondence algorithm and the Hausdorff distance.

6. Method according to one of the preceding claims, in which data from vehicles belonging to said user (GAR) are acquired, and said succession of steps (SET) is adapted as a function of said data from vehicles belonging to said user.

7. Method according to one of the preceding claims, wherein said models are trained from time-stamped learning geolocation data (DAP) by machine learning using at least one neural network.

8. Method according to one of the preceding claims, in which said succession of steps (SET) first implements the determination of a motorized or non-motorized mode of transport (NMO).

9. Method according to one of the preceding claims, in which said succession of steps (SET) implements the determination of a mode of transport by motorcycle (MOT) before the determination of a mode of transport by public transport not circulating on a road network (TRC) and before determining a mode of transport circulating on a road network (VER).

10. Method according to one of the preceding claims, in which said succession of steps (SET) implements the determination of a mode of public transport (TRC) not circulating on a road network before the determination of a mode of transport circulating on a road network (VER).

11. Method according to one of the preceding claims, wherein said heuristic filter is based on a statistical analysis of said time-stamped learning geolocation data (DAP).

12. Method according to one of the preceding claims, in which a mapping algorithm is applied per mode of public transport (TRC) not circulating on a road network.

13. A method according to one of the preceding claims, wherein if the mode of transport is determined to be non-motorized, said trained model and / or said heuristic filter further determines the type of non-motorized transport.

Images