WO2019197345A1

WO2019197345A1 - Vehicular motion assessment method

Info

Publication number: WO2019197345A1
Application number: PCT/EP2019/058824
Authority: WO
Inventors: German CASTIGNANI; Leandro MASELLO
Original assignee: Motion-S
Priority date: 2018-04-09
Filing date: 2019-04-08
Publication date: 2019-10-17
Also published as: EP3774478A1; LU100760B1

Abstract

A vehicular motion assessment method (Fig.1) comprises: capturing motion observations on-board a vehicle(12), the motion observations including timestamped vehicle positions; logging the motion observations; map-matching the motion observations so as to identify, for each vehicle position a road link of a road network stored in a map database (18) as well as a map-matched position on the road link; contextualizing the motion observations by augmenting them with situational data obtained by querying the map database (18)or another geographic information database; determining, based on the contextualized motion observations, whether the vehicle (12) was exposed, during the trips, to unsafe driving situations; and communicating one or more risk scores indicative of accident probability, calculated based on determined exposures to unsafe driving situations.The one or more risk scores indicative of accident probability are scaled by spatial and/or temporal exposure factors, the spatial and/or temporal exposure factors being indicative of exposure to risk due to geographical and/or temporal circumstances of the unsafe driving situations.

Description

VEHICULAR MOTION ASSESSMENT METHOD Field of the Invention

[0001] The invention generally relates to monitoring vehicular motion and driving style profiling by assessing exposure of the vehicle to unsafe driving situations, in particular unsafe driving situations imputable to the driver.

Background of the Invention

[0002] Monitoring driving activity has become an important topic in the recent years. Two main application fields have been devised in which real-time analysis and evaluation of the driving behaviour will play a prominent role. The first application field is (car) fleet management, where real-time information on the driving behaviour may be used in taking measures aiming at motivating the drivers to adopt a more efficient driving style, to reduce vehicle wear, to reduce their energy footprint and to save fuel. On the scale of the fleets, significant savings in terms of CO2 emissions and fleet depreciation may be achieved. The second important application field is the car insurance market. The calculation of car insurance premium may be enriched by or exclusively be based on individual parameters like the covered distance, the time spent on the road, the time of day, the roads used, etc. Those concepts are usually referred to as Pay-As-You-Drive (PAYD), Pay-How-You-Drive (PHYD) or Usage-Based Insurance (UBI). An important parameter in such premium calculation schemes is the driving style (i.e. an indicator of how many and how important are risks the driver takes when diving). Energy consumption, vehicle wear and insurance being the most relevant cost factors influencing the total cost of mobility, such applications could be used as the main pillars for enabling dynamic pricing in mobility, in particular for new mobility means like shared mobility offers.

[0003] International patent application WO 2013/104805 A1 discloses an apparatus, a system and a method for calculating a driving behaviour risk indicator for a vehicle driver. The system uses a 3D inertial sensor with 3D gyroscope functionality and the risk indicator is based on events detected using the sensor, in particular on the count of events occurring in predetermined categories associated with dangerous and/or aggressive driving. [0004] International patent application WO2016/107876 A1 discloses a vehicular motion monitoring method, wherein motion observations are captured on-board a vehicle with one or more sensors and sets of motion observations are mapped onto a respective feature vector in an at least two-dimensional feature space, each feature vector having a first vector component representative of a longitudinal motion characteristic and a second vector component representative of a lateral motion characteristic. Determinative parameters of a multivariate Gaussian probability density function modelling a population of collected feature vectors are then updated and a riskiness indicator is assigned to each feature vector. The calculation of the riskiness indicator is based upon an event severity indicator indicative of how anomalous each feature vector is in comparison to the modelled population and upon a position of the feature vector relative to one or more previous and/or subsequent feature vectors. The riskiness indicator is integrated over time so as to obtain a risk assessment of the driving style. The method described in WO2016/107876 A1 suffers from several drawbacks, in particular, from the fact that the telematics metrics (acceleration, braking, cornering and speed) have no objective connection to risk statistics.

Summary of the Invention

[0005] A first aspect of the invention relates to a vehicular motion assessment method, comprising:

a. capturing motion observations with a mobile terminal on-board a vehicle, the mobile terminal including or being connected to a GNSS receiver, the motion observations including timestamped vehicle positions (“trackpoints”) measured by the GNSS receiver;

b. logging the motion observations;

c. map-matching the motion observations so as to identify, for each vehicle position a road link of a road network stored in a map database as well as a map-matched position on the road link;

d. contextualizing the motion observations by augmenting them with situational data obtained by querying the map database and/or another geographic information database (e.g. a weather database); e. determining, based on the contextualized motion observations, whether the vehicle was exposed, during the trips, to unsafe driving situations, e.g. unsafe driving situations imputable to the driver; and

f. communicating (e.g. via a display and/or via a telecommunication link) one or more risk scores indicative of accident probability, calculated based on determined exposures to unsafe driving situations.

[0006] The vehicular motion assessment method may be carried out by the mobile terminal, which could comprise a smartphone, a tablet, a phablet, a notebook, an on- board diagnostic unit, a dedicated data logger, or the like. The mobile terminal could be separate from or integrated into the vehicle. The map database and/or the other geographic information database could be stored on the mobile terminal but it could also be located elsewhere, with the mobile terminal having access to it via a private network or the Internet.

[0007] Those skilled in the art will appreciate that certain steps of the vehicular motion assessment method could be carried out remotely from the mobile terminal. Accordingly, a second aspect of the invention relates to a vehicular motion assessment method, wherein above steps c., d., e. and f. are delegated to a data collection platform (also called“contextual ization and profiling platform”). The method according to the second aspect thus comprises above steps a. and b., and further:

o establishing a secure communication link with a data collection platform configured to carry out above steps c., d., e. and f.; and

o transmitting the logged motion observations via the communication link to the data collection platform.

In order to allow unambiguous identification of the vehicle by the data collection platform, the mobile terminal establishes a connection with the vehicle and tags the motion observations captured while the data connection exists with an identifier of the mobile terminal and/or of the vehicle known to the data collection platform. The connection the mobile terminal establishes with the vehicle could be a data connection. However, although the possibility of data exchange between the vehicle and the mobile terminal may be preferable, it is not necessary for all applications of the invention. For instance, the connection could be established via a plug with wireless functionality that connects to the vehicle power supply (e.g. an ISO 4165:2001 connector): when the plug is powered by the vehicle, the wireless functionality is on and the mobile terminal may sense the presence of the plug and hence that the vehicle power supply is on.

[0008] A third aspect of the invention relates to a vehicular motion assessment method capable of being carried out by a data collection platform upon receipt of the motion observations. The method according to the third aspect comprises: o obtaining (e.g. via a data link, from a computer memory, etc.) motion observations logged on-board a vehicle, the motion observations including timestamped vehicle positions; o determining if the obtained motion observations are organized in trips and, if they are not, dividing them into trips by detecting spatial and/or temporal discontinuities and/or a driver change in the obtained motion observations; o map-matching the motion observations so as to identify, for each vehicle position a road link of a road network stored in a map database as well as a map-matched position on the road link; o contextualizing the motion observations by augmenting them with situational data obtained by querying the map database and/or another geographic information database; o determining, based on the contextualized motion observations, whether the vehicle was exposed, during the trips, to unsafe driving situations; and o communicating (e.g. to the mobile terminal or an authorized third party) one or more risk scores indicative of accident probability, calculated based on determined exposures to unsafe driving situations.

[0009] Preferred embodiments of the different aspects of the invention are discussed below. The features of these preferred embodiments may apply to all aspects of the invention unless it is clear from the context that this is not the case.

[0010] The motion observations could be logged with a frequency of at least 0.2 Hz (corresponding to a trackpoint every 5 s). Higher frequencies may be preferred, e.g. 0.33 Hz, 0.5 Hz, 1 Hz, or still higher. The frequency with which the motion observations are logged could also be dynamically adjusted depending on vehicle acceleration (including longitudinal and lateral acceleration of the vehicle). [0011] The motion observations could include data collected by on-board sensors, e.g., information from smart tyres, shifting patterns, rotational speed, etc. Such vehicle data are preferably retrieved from the vehicle’s onboard communication system(s). Any data connection between the mobile terminal and the vehicle (e.g. the vehicle’s internal communications network, the CAN bus, a wireless communications interface of the vehicles, or the like) could be established over a wire connection or via a wireless link (e.g. Bluetooth™). Successful establishment of the data connection with the vehicle could trigger the capturing of the motion observations and the rupture of the data connection with the vehicle could terminate the capturing of the motion observations.

[0012] Preferably, the motion observations are organized in trips, and a unique trip identifier is associated with the motion observations captured during a continuous time interval starting with successful establishment of the (data) connection with the vehicle and ending with the rupture of the (data) connection. In other words, as long as a (data) connection between the mobile terminal and the vehicle uninterruptedly exists all motion observations captured are associated with the same unique trip identifier. In this context, it may be worthwhile noting that the trip identifier and the vehicle and/or mobile terminal identifier could be combined into a complex identifier containing both pieces of information.

[0013] The motion observations could be transmitted on-the-fly to the data collection platform. Alternatively or additionally, the mobile terminal could buffer the logged motion observations in a memory and transmit the buffered motion observations batch- wise to the data collection platform. The mobile terminal may be configured such that it buffers the logged motion observations only when on-the-fly transmission is not possible. Alternatively, buffered transmission could be the default option.

[0014] Preferably, the motion observations include, for each timestamped vehicle position, at least one of vehicle speed, vehicle heading, vehicle acceleration, rate of turn and engine rotational speed. When vehicle speed, vehicle heading and vehicle acceleration are not part of the raw data, they can be computed based on the logged motion observations. The determination whether the vehicle was exposed to unsafe driving situations is preferably also based on vehicle speed, vehicle heading and vehicle acceleration.

[0015] As used herein, the expression“situational data” refers to data correlating, in terms of time and position, with the motion observations, so that they describe specific aspects of the situation in which the motion observations were made. The situational data used for augmenting the motion observations could include speed limits, road topology (e.g. slope, curvature), road type (e.g. motorway, rural road, urban road), number of lanes, traffic signs and traffic information (e.g. real-time traffic data or historical traffic data, including for instance, average speed of vehicles on a certain road link for a certain date and time, etc.) The situational data could also include meteorological data obtained by querying a meteorological database. The meteorological database could be separate from the map database of integrated therein, e.g. as one or more“layers”.

[0016] Exposure of the vehicle to unsafe driving situations could be tested by processing a list of unsafe driving situations statistically correlating with road accidents and determining based on the contextualized motion observations whether and how often a listed unsafe driving situation occurred. In practice, such a list may be implemented as a series of tests and/or one or more decision trees.

[0017] The listed unsafe driving situations could include:

o breaches of traffic regulations (such as, e.g., speeding, illegal turns, illegal U- turns, running a stop sign, etc.) and

o potentially dangerous manoeuvers or driver behaviour (such as, e.g., high braking, high steering, harsh acceleration, fatigue driving, driving faster than the traffic flow, etc.)

[0018] In the calculation of the risk scores, each exposure to an unsafe driving situation is preferably accounted for according to the statistical probability of this unsafe driving situation contributing to accidents. This statistical probability may be determined on the basis of road safety statistics and be stored as a parameter in the software. In the calculation of the risk scores, the severity of the accidents (e.g. fatal, serious, slight) may be taken into account by weighting each exposure to an unsafe driving situation according to the statistical probabilities of this unsafe driving situation contributing to accidents of the different severities.

[0019] In particular, it will be appreciated that the vehicular motion assessment method may include determining contextual characteristics of the locations and the time the vehicle was moving (e.g. as part of the contextual ization step or the step of detection of unsafe driving situations or as a separate step), and further determining an exposure to risk, based on past accident statistics at similar times and in roads with similar contextual characteristics.

[0020] Preferably, the one or more risk scores indicative of accident probability are scaled by spatial and/or temporal exposure factors that are indicative of exposure to risk due to geographical and/or temporal circumstances of the unsafe driving situations. For instance, the spatial and/or temporal exposure factors could be indicative of exposure to risk considering geographical accident density and temporal accident density with respect to time of day, respectively. The spatial and/or temporal exposure factors could also be indicative of exposure to risk considering the similarity, in terms of road characteristics and/or other contextual information, to real reported accidents, It is worthwhile noting that the spatial and/or temporal exposure factors may be calculated taking historical accident data into account. In particular, the geographical accident density could be based on the locations of accidents reported during a past reference period. Likewise, the temporal accident density (number of accidents per time unit as a function of time of day) could be based on statistical historical data collected during a reference period. The spatial and/or temporal exposure factors could be broken down according to the severity of accidents (e.g. fatal, serious, slight). The spatial and temporal exposure factors could be treated as independent parameters but it would also be possible to use spatial exposure factors broken down according to the time of day, day of the week, and/or other parameters.

[0021 ] When the spatial and/or temporal exposure factors relating to a given situation, which is identified only by its time and location, are derived from historical geographical and temporal accident densities, accident and road safety data need be available in sufficient quantity for all geographic regions or countries where the vehicular motion assessment service is offered. This approach thus has the drawback that geographically incomplete accident and road safety data may lead to“blind spots”, where spatial and/or temporal exposure factors cannot be calculated (and thus has to be set to a default value, e.g. unity).

[0022] According to a more sophisticated approach, the available accident and road safety data are processed to extract the corresponding spatiotemporal features (in terms of, e.g., road topology, whether, traffic density, daylight, etc.) The historical accident and road safety data may then serve as training data in a machine learning algorithm programmed to map a co-occurrence of spatiotemporal features (“feature vector”) on corresponding spatial and/or temporal exposure factors that are indicative of exposure to risk due to geographical and/or temporal circumstances. At runtime, unsafe driving situations are processed in order to determine the corresponding feature vectors and the spatial and/or temporal exposure factors. This approach thus identifies, for a given driving situation, those spatiotemporal features that have been observed to be correlated to accidents in the historical accident and road safety data and calculates the spatial and/or temporal exposure factors using the model obtained in the training step. It should be noted that the machine learning algorithm could be configured to learn the relevant spatiotemporal features by itself. Preferably, the training is complemented when new accident and road safety data become available in order to improve the model. Due to the fact that this approach is not based only on the location and time information of past accidents but looks for similarities and patterns, the above- mentioned risk of“blind spots” is significantly mitigated, if not excluded.

[0023] The one or more risk scores indicative of accident probability are preferably obtained by scaling exposures to unsafe driving situations imputable (solely) to the driver due to their behaviour with the spatial and/or temporal exposure factors. In other words, prior to the scaling, the exposures to unsafe driving situations are preferably measured without regard to the location and/or to the time of day of their occurrence but taking into account detected exposure to unsafe driving situations imputable (solely) to the driver.

[0024] The contextual ization of the motion observations could further comprise the determination of adverse circumstances (requiring the driver to exercise particular caution), e.g. rain, fog, driving against the sun, high traffic density, etc. Unsafe driving situations imputable to the driver occurring in such adverse circumstances could be weighted more highly in calculation of the one or more risk scores indicative of accident probability (if accident statistics corroborate such higher weighting). Certain adverse circumstances may be directly available among the contextual ization data. Others could be derived by combining directly available motion observations and/or contextual ization data. For instance, when the vehicle is driving under a clear sky (derivable from meteorological data) on a motorway (derivable from the“road type” information), glare by the sun could be detected by inputting vehicle position and heading, date and time into a sun position calculator. [0025] Preferably, in the calculation of the one or more risk scores indicative of accident probability, unsafe driving situations imputable to the driver are taken into account depending on the travelled distance over which the unsafe driving situations existed.

[0026] It may be worthwhile noting that the present invention makes possible more targeted education of drivers having a risky driving style. The communication of the risk scores could be made to the driver (e.g. via the mobile terminal) and could include recommendations for reducing the riskiness of their driving style based on the determined unsafe driving situations imputable to the driver. For instance, if it is detected that a driver takes turns with relatively high speed, the driver could be informed about how much their risk score (and, possibly, their insurance premium) could be reduced if they adopted a more prudent driving style.

Brief Description of the Drawings

[0027] The accompanying drawing illustrates several aspects of the present invention and, together with the detailed description, serves to explain the principles thereof. In the drawings:

Fig. 1 : is a schematic illustration of an embodiment of a vehicular motion assessment method in accordance with the invention.

Detailed Description of a Preferred Embodiment

[0028] The proliferation of mobile sensing and telematics systems in connected vehicles have enabled the big-data industry in mobility. In particular driver profiling for insurance purposes arises as a must-have for setting up UBI systems. In the preferred embodiment of the invention discussed below, driver profile correlation to objective driver risk is presented. It should be noted, however, that the invention is not limited to this application, which has been chosen only for the purpose of illustration. Another possible application of the invention could, for instance, be predictive maintenance.

[0029] According to the preferred embodiment, objective risk scores are provided based on behavioural, spatial and temporal factors computed over contextualized vehicular motion observations, which may be as simple as GNSS (Global Navigation Satellite System) tracks and/or contain further data measurable within the vehicle (e.g. speed, acceleration, vibrations, etc.) To compute risk scores, the probabilities of these factors being observed in real accidents, which are extracted from publicly available road safety statistics, serve as the empirical, objective basis.

[0030] There is a need to understand how driving behaviour variables impact the driver’s risk of having an accident. Such understanding could be obtained by correlation analysis of driving data (i.e., telematics data) and real insurance claims. However, the required data are not readily available and even less so to the broader public. Specifically, generating such a dataset would require a long time and important investments. In the context of the present embodiment of the invention, historical road safety statistics are used to circumvent the above difficulty. Road safety reports present important amounts of information about accidents. To establish a link between vehicular motion observations and accident risk, reports dealing with the main causes of accidents have been considered. Specifically, a contextual ization of the raw motion observations is carried out. The contextual ization comprises augmenting the raw motion observations (i.e. the measurements made in the vehicle) with map data (situational data) obtained by querying a map database (e.g. HERE, Google Maps, OpenStreetMap) and extracting a set of risk-based telematics variables from the contextualized vehicular motion observations.

[0031] Prior to analyzing motion observations and determining risk scores, a list of telematics metrics or“unsafe driving situations” is determined that have well defined and documented links to risk. This list of telematics metrics (TM) can be computed in advance and the rules to compute their values may then be stored in the software. These rules need not be updated frequently but updates may be advisable if new road safety reports become available of if new TM are to be computed.

[0032] A list of TM examples used in the context of the present embodiment of the invention is provided in table 1 below. It should be noted that this list is not exhaustive and could be extended (or reduced) in accordance with the needs of the specific application.

Table 1

[0033] Of the listed TM, TM1 -TM18 as well as TM21 -TM24 and TM35-TM36 represent unsafe driving situations imputable to the driver (behavioural TM). TM19 and TM20 depend on the types of roads driven on. TM25 to TM30 are spatial (TM25-TM27) and temporal (TM28-TM30) exposure factors indicative of exposure to risk considering geographical accident density and temporal accident density with respect to time of day, respectively.

[0034] The inventors identified the above listed TM by studying the Contributory Factors (CF) to past road accidents. Specifically, in the context of the present embodiment, the inventors adopted the methodology proposed by the UK Department for Transport (https://www.qov.uk/qovernment/statistical-data-sets/ras50-contributorv- fact rs). CF represent a measure of the main causes that led directly to an accident of a given severity (fatal, serious and slight). On every accident reported, the authorities (e.g., police officer) fill a form (called STATS19) where the CF observed in the accident are marked. A detailed explanation of each individual CF is given in the STATS20 document (“Instructions for the Completion of Road Accident Reports from non- CRASFI Sources”, available online under https://www.qov.uk/qovernment/uploads/svstem/uploads/attachment data/fiie/23059 6/stats20-2011.pdf). The above official documents are mentioned for the purpose of illustration. Any other road safety report that follows a similar CF approach could be used in the context of the present invention, which is not limited to the use of the above- mentioned road safety report.

[0035] Out of the 78 CF presented in STATS19, a subset of at least 21 CF that can be measured through TM was identified. Table 2 below indicates the 22 identified CF measurable with TM. Again, it should be noted that Table 2 is not exhaustive and that the used CFs could be adapted depending on the needs of the application. Table 2

[0036] The proposed methodology aims at predicting, for every trip, which of those CF would have been triggered in a hypothetical accident. This represents an a priori judgement of the driver behaviour during the trip by using the same methodology as for police reported accident.

[0037] In order to do so, each measurable CF is computed as a binary variable (i.e., enabled or disabled) using a logical combination of TM constrained by predefined thresholds. This combination is done in a way to well-represent the CF involved. For instance, as illustrated in Table 2, CF 7,’’Exceeding Speed Limit”, is computed using TM1 (Speeding average) and TM2 (Speeding proportion). Also, CF 16,“Aggressive driving”, takes into account TM8 (High steering), TM10 (Positive Kinetic Energy) and TM11 (Negative Kinetic Energy). These metrics have been proven to be highly correlated to aggressive driving in the literature, see G. Castignani, T. Derrmann, R. Frank, and T. Engel,“Validation study of risky event classification using driving pattern factors,” in 2015 IEEE Symposium on Communications and Vehicular Technology in the Benelux (SCVT), pp. 1-6, Nov. 2015 and D. A. Johnson and M. M. Trivedi,“Driving style recognition using a smartphone as a sensor platform,” in 2011 14th International IEEE Conference on Intelligent Transportation Systems (ITSC), pp. 1609- 1615, Oct. 2011. More complex CF, like CF 3, ’’Disobeyed Give Way or Stop”, needs consideration of the traffic signs information coming from the contextual ization process, in combination with TM14 and TM15, which measure the driver’s attitude while approaching yield and stop signs.

[0038] In the present embodiment of the invention, in order to compute an objective measure representing the vehicle’s exposure to risk (i.e. a risk score), the CF presented above are linked with their corresponding occurrences in accidents given by statistical reports that show information about the presence of CF in accidents according to their severity (i.e. fatal, serious, slight) and to the road type where they occurred (i.e. motorway, urban and rural roads). This calculation is performed for every trip of a vehicle in accordance with Eq. 1 :

In Eq. 1 , N is the number of CF and a_c. is a binary variable which represents whether the i^th CF has been enabled in the trip or not. As stated above, this is measured using a combination of TM (see Table 2). If a given CF has been observed in the trip, the exposure for that CF is computed using its related statistics. Firstly, the exposure according to the road type j (j = 1 , M) is computed using W_Rj, which is the probability of the i^th CF being involved in an accident on a road of type R_j. M could be 3 with j = 1 , 2 or 3 meaning motorway, urban or rural road, respectively. d_Rj is the proportion of distance driven on each road type j. Then, this value is used along with the statistics regarding accident severity, where

are the probabilities of the i^th CF being involved in fatal, serious and slight accidents, respectively. Finally, the sum is computed over all CF, giving as a result the vector s_B = (s , s|^e, s|^J )^T (superscript“T” meaning transposition), which represents the behavioural (i.e. driver-imputable) exposure to fatal, serious and slight accidents for the given trip. It is worthwhile noting that Eq. 1 could be easily adapted if the accident categories (here: fatal, severe and slight) changed.

[0039] The exposures are then combined in order to get a risk score. This can be done by using accident cost information from road safety reports (e.g.“Accident and casualty costs (RAS60)” available under https://www.gov.uk/government/statistical· data-sets/ras60-averaqe-value-of-preventinq-road-accidents). Specifically, in this example, we compute C = (C^F , C^Se, C^sl )^T , representing the average normalized cost of fatal, serious and slight accidents. Then, the risk score of a trip, R^F _rip that takes into account behavioural exposure only may be calculated by:

(Eq. 2) where“·” represents the scalar product.

[0040] Parameters W_p ^l, w

(i = 1 , ... , 17) as well as C^F, C^Se, C^sl are calculated in advance. Like the rules for computing the TM, these parameters need not be changed frequently but updates can be made at larger intervals (e.g. every year).

[0041] Factors other than behavioural ones may have an impact on the risk score. In particular, the route taken as well as the time and length of the trip might increase or decrease the potential exposure to have an accident. By using the history of accidents reported, R^F _rip can be extended using spatial and temporal scaling to provide the final metric R_Trip. This may be done by defining diagonal matrices (see Eq. 3 and Eq. 4 below), where each component is computed using TM25 to TM30 defined in Table 1 and computing R_Trip as indicated in Eq. 6:

[0042] The diagonal components of S_spatial and S_temporal preferably follow exponential distributions with values ranging from 0.5 to 1.5. This allows obtaining an overall trip risk score R_Trip that is expressed as a scaling of the behavioural factor with spatial and temporal weighting factors. Therefore, R^_rip is scaled up or down according to the individual route, time and length of the trip.

[0043] For the spatial weighting matrix S_spatia road segments (i.e. road links of the map database) where accidents of different severities have been reported are taken into account. For the temporal weighting matric S_temporai the time slots of those accidents are taken into account. In other words, the spatial and temporal weighting matrices depend on the geographical local density of past accidents and on the temporal density, with respect to time of day, respectively. The effect of the spatial and temporal weighting matrices may be illustrated by considering two drivers having the same behavioural exposure s_B and driving trips of equal length. Supposing that the trips take place simultaneously (resulting in identical S_temporai matrices for both drivers), the driver who drives on road segments where accidents have been reported in the past will obtain a higher risk score than his fellow who drives on road segments where no accidents have happened. Supposing that the trips take place on the same roads (resulting in identical S_spatiai matrices for both trips) but at different times (e.g. on different days of the week or at different times of the day), the driver who drives at a time at which accidents are more frequent will obtain a higher risk score than the driver who drives at a time at which accidents are less frequent.

[0044] As an alternative to separate (or separable) matrices S_spatial and S_temporal, one could use a joint spatiotemporal weighting factor, S_{spatio -temporal}, such that R_Trip can be computed as indicated in Eq. 7. Such a joint spatiotemporal weighting factor could come from a machine-learning model, including algorithms for clustering and classification that can infer the likelihood of a driver being exposed to similar characteristics observed in fatal, serious and slight accidents.

[0045] Fig. 1 summarizes the overall process according to the illustrative embodiment of the invention. Illustrations of the different steps are provided on the left-hand side and the corresponding data made available are shown in the test boxes on the right- hand side.

[0046] Motion observations are recorded by a mobile terminal 10 on-board a vehicle 12. The mobile terminal 10 is in this case represented as a smartphone (for the purpose of illustration). The mobile terminal 10 includes a GNSS receiver or is connected with the GNSS receiver of the vehicle 12 and logs the vehicle position with a frequency of, e.g., 1 Hz. The recording starts when the mobile terminal has established a data connection 13 (e.g. a Bluetooth™ connection) with the vehicle’s wireless communication interface 14. Optionally, an additional condition for the start of the recording could be that an initial position change is detected. As long as the data link exists without interruption, all motion observations are deemed to belong to the same trip and are stored under the same trip identifier (tripID). The recording ends when the data connection 13 is terminated. The corresponding trip is closed in consequence and the next trip will be given another TripID. In the drawing, the raw motion observations are represented by a file having a header with the vehicle identifier (vehiclelD) and the tripID and a matrix wherein each line corresponds to one recording instant.

[0047] Regarding the representation in matrix form it should be noted that this representation was chosen because it is convenient for illustration but is not meant to imply any particular data structure. For instance, the motion observations could be recorded in an XML file. In the illustrated embodiment, each data point of the raw motion observations includes the time (ί_έ), the position ( _£), the velocity {v_t) obtained from the GNSS receiver, and, possibly, readings from other sensors (not shown in the drawing), e.g. accelerations and turn rates obtained by an IMU. It may be worthwhile noting that the file could contain other metadata, such as, e.g. identifiers of the mobile terminal, the driver, the (type of) GNSS receiver, etc.

[0048] The mobile terminal 10 transmits the motion observations to a data collection platform 16 via a secure communication link. The management of the secure communication link is preferably one of the tasks carried out by the application (“app”) that also manages data connection with the vehicle and the collection of the motion observations.

[0049] The data collection platform 16 is connected to a map database 18, which it queries in order to map-match the raw motion observations. The map-matching step comprises finding the road links (of the map) that correspond to the measured positions. Due to possible measurement errors, some measured positions could lie off- road. In this case, the map-matching step returns the road links (in the form of the corresponding road link identifier “UnkID ") and corrected positions (x_corr , which satisfies the constraint that it lies on the i^th road link of the map) that best fit with the raw motion observations. The data collection platform 16 further augments the motion observations with map data retrieved from the map database. These additional map data (contextual ization data) serve to explain the situation in which the motion observations were made, which allows assessing the motion observations in their context. Contextual ization data include, for instance, the speed limits {vjnax^ in force on the road links, road type, (historic) traffic information (local traffic density, local average speed at the time of the observations), information on traffic signs (e.g.“no overtaking”,“no U-turn”,“yield”,“stop”,“pedestrian crossing”,“slippery road”, etc.), historic weather data, information on past geographical and temporal accident density. In order not to overload the drawing, only the speed limit information is explicitly shown thereon.

[0050] The next step of the process comprises the analysis of the contextualized motion observations (“profiling”). That analysis involves computing the telematics metrics (TM1 -TM30, cf. Table 1 ) for each trip. Specifically, the data collection platform 16 processes the list of telematics metrics TM1 -TM30 (which are linked to Contribution

Factors statistically correlating with accident probabilities, cf. Table 2) and determines the value of each telematics metrics for the trip under examination. The analysis step further comprises determining the different risk scores. In the illustrated example, s_B and R_Trip are determined via Equations 1 and 5. Finally, risk integration is carried out: this step comprises calculating an overall risk score R_overau taking into account all trips of a given period. [0051] Finally, the different risk scores are communicated to the customers. In the illustrated case, these include the driver, who receives his scores on their smart phone 10, a third party 20 or both. The third party could be a fleet management company or an insurer. It should be noted that the data communicated to third parties could be anonymized, if necessary.

[0052] It is worthwhile noting that the data collection platform could be implemented as a cloud computing platform. Preferably, it is divided into different functional blocks, comprising, in particular, a map-matching and contextual ization block 22 and a profiling block 24. [0053] Risk-scores could also be used for coaching purposes. Drivers with risky driving behaviour could be intelligently guided with driving advice related to their profile. Beyond insurance applications, road safety associations and authorities could set up driving campaigns based on the proposed risk score to increase awareness in road security and safety.

[0054] While a specific embodiment and a specific example have been described herein in detail, those skilled in the art will appreciate that various modifications and alternatives to those details could be developed in light of the overall teachings of the disclosure. Accordingly, the particular arrangements disclosed are meant to be illustrative only and not limiting as to the scope of the invention, which is to be given the full breadth of the appended claims and any and all equivalents thereof.

Claims

1. A vehicular motion assessment method, comprising:

a. capturing motion observations with a mobile terminal on-board a vehicle, the mobile terminal including or being connected to a GNSS receiver, the motion observations including timestamped vehicle positions measured by the GNSS receiver during a trip;

b. logging the motion observations;

d. contextualizing the motion observations by augmenting them with situational data obtained by querying the map database and/or another geographic information database;

e. determining, based on the contextualized motion observations, whether the vehicle was exposed, during the trip, to unsafe driving situations; and f. communicating one or more risk scores indicative of accident probability, calculated based on determined exposures to unsafe driving situations, wherein the one or more risk scores indicative of accident probability are scaled by spatial and/or temporal exposure factors, the spatial and/or temporal exposure factors being indicative of exposure to risk due to geographical and/or temporal circumstances of the unsafe driving situations.

2. A vehicular motion assessment method, comprising:

capturing motion observations with a mobile terminal on-board a vehicle, the mobile terminal including or being connected to a GNSS receiver, the motion observations including timestamped vehicle positions measured by the GNSS receiver during a trip;

logging the motion observations;

establishing a secure communication link with a data collection platform configured to carry out steps c., d., e. and f. defined in claim 1 ;

transmitting the logged motion observations via the communication link;

wherein the mobile terminal establishes a connection with the vehicle and tags the motion observations captured while said connection exists with an identifier of the mobile terminal and/or of the vehicle known to the data collection platform.

3. The vehicular motion assessment method as claimed in claim 2, wherein successful establishment of the connection with the vehicle triggers the capturing of the motion observations, wherein the rupture of the connection with the vehicle terminates the capturing of the motion observations.

4. The vehicular motion monitoring method as claimed in claim 2 or 3, wherein the connection the mobile terminal establishes with the vehicle is a data connection.

5. The vehicular motion assessment method as claimed in any one of claims 1 to 4, wherein the motion observations are organised in trips, and wherein a unique trip identifier is associated with the motion observations captured during a continuous time interval starting with successful establishment of the data connection with the vehicle and ending with the rupture of the data connection.

6. The vehicular motion assessment method as claimed in any one of claims 2 to 5, wherein the motion observations are transmitted on-the-fly to the data collection platform.

7. The vehicular motion assessment method as claimed in any one of claims 2 to 6, wherein the mobile terminal buffers the logged motion observations in a memory and transmits the buffered motion observations batchwise to the data collection platform.

8. The vehicular motion assessment method as claimed in any one of claims 1 to 7, wherein the motion observations include, for each timestamped vehicle position, at least one of vehicle speed, vehicle heading, vehicle acceleration, rate of turn and engine rotational speed.

9. The vehicular motion assessment method as claimed in any one of claims 1 to 8, wherein the motion observations are logged with a frequency of at least 0.2 Hz during parts of the trip.

10. The vehicular motion assessment method as claimed in any one of claims 1 to 9, wherein the frequency with which the motion observations are logged is dynamically adjusted depending on vehicle acceleration.

11. The vehicular motion assessment method as claimed in any one of claims 1 to 10, wherein the one or more risk scores indicative of accident probability are obtained by scaling exposures to unsafe driving situations imputable to the driver due to their behaviour with said spatial and/or temporal exposure factors.

12. Vehicular motion assessment method, comprising:

obtaining motion observations logged on-board a vehicle by a mobile terminal, the motion observations including timestamped vehicle positions;

determining if the obtained motion observations are organised in trips and, if they are not, dividing them into trips by detecting spatial and/or temporal discontinuities and/or a driver change in the obtained motion observations;

map-matching the motion observations so as to identify, for each vehicle position a road link of a road network stored in a map database as well as a map-matched position on the road link;

contextualizing the motion observations by augmenting them with situational data obtained by querying the map database or another geographic information database;

determining, based on the contextualized motion observations, whether the vehicle was exposed, during the trips, to unsafe driving situations; and communicating one or more risk scores indicative of accident probability, calculated based on determined exposures to unsafe driving situations, wherein the one or more risk scores indicative of accident probability are scaled by spatial and/or temporal exposure factors, the spatial and/or temporal exposure factors being indicative of exposure to risk due to geographical and/or temporal circumstances of the unsafe driving situations.

13. The vehicular motion assessment method as claimed in any one of claims 1 to

12, wherein the situational data used for augmenting the motion observations include speed limits, road topology, road typeclass, number of lanes, traffic signs and/or traffic information.

14. The vehicular motion assessment method as claimed in any one of claims 1 to

13, wherein the situational data used for augmenting the motion observations include meteorological data.

15. The vehicular motion assessment method as claimed in any one of claims 1 to

14, wherein the logged motion observations include vehicle speed, vehicle heading and vehicle acceleration or wherein vehicle speed, vehicle heading and vehicle acceleration are computed based on the logged motion observations, and wherein the determination whether the vehicle was exposed to unsafe driving situations is also based on vehicle speed, vehicle heading and vehicle acceleration.

16. The vehicular motion assessment method as claimed in any one of claims 1 to 15, wherein the logged motion observations include data collected by on-board sensors, e.g., information from smart tyres, shifting patterns, rotational speed, etc.

17. The vehicular motion assessment method as claimed in any one of claims 1 to 16, wherein exposure of the vehicle to unsafe driving situations is tested by processing a list of unsafe driving situations statistically correlating with road accidents and determining based on the contextualized motion observations whether and how often a listed unsafe driving situation occurred.

18. The vehicular motion assessment method as claimed in claim 17, wherein the list of unsafe driving situations includes unsafe driving situations imputable to the driver.

19. The vehicular motion assessment method as claimed in claim 17 or 18, wherein the listed unsafe driving situations include:

o breaches of traffic regulations (such as, e.g., speeding, illegal turns, illegal U-turns, running a stop sign, etc.) and

o potentially dangerous manoeuvers or driver behaviour (such as, e.g., high braking, high steering, harsh acceleration, fatigue driving, etc.)

20. The vehicular motion assessment method as claimed in any one of claims 1 to

19, wherein the one or more risk scores indicative of accident probability are scaled by spatial and/or temporal exposure factors, the spatial exposure factors being indicative of exposure to risk considering geographical accident density, the temporal exposure factors being indicative of exposure to risk considering temporal accident density with respect to time of day.

21. The vehicular motion assessment method as claimed in any one of claims 1 to

20, wherein the contextual ization of the motion observations comprises the determination of adverse circumstances and wherein unsafe driving situations imputable to the driver having occurred in such adverse circumstances are weighted more highly in calculation of the one or more risk scores indicative of accident probability.

22. The vehicular motion assessment method as claimed in any one of claims 1 to 21 , wherein, in calculation of the one or more risk scores indicative of accident probability, unsafe driving situations are taken into account depending on the travelled distance over which the unsafe driving situations existed.

23. Application comprising instructions, which, when executed by a processor of a mobile terminal, such as, e.g. a smartphone, a tablet, a phablet, a notebook, an on-board diagnostic unit or the like, cause the mobile electronic device to carry out the method as claimed in any one of claims 1 to 22.