WO2014108219A1 - Apparatus, system and method for vehicle trajectory reconstruction - Google Patents

Abstract

An apparatus for use in reconstructing a vehicle trajectory, comprises: a processing and control unit; and a memory, the apparatus being adapted to: capture inertial data at a predetermined frequency from an inertial unit mounted on the vehicle, the inertial unit including a 3D inertial sensor with 3D gyroscope functionality; update a sensor error model based on the received inertial data and store the updated sensor error model; detect an event; and after the event is detected, store the captured inertial data until a predetermined time after the start of the event.

Description

APPARATUS, SYSTEM AND METHOD

FOR VEHICLE TRAJECTORY RECONSTRUCTION

Field of the Invention

The present invention is related to a technique for reconstructing the trajectory of a vehicle, for example after a crash.

Background of the Invention

Satellite navigation systems for determining the position of and navigation of a vehicle are well-known. The use of event data recorders (EDRs) and accident detectors comprising accelerometers is also known. Accordingly, rich data from sources such as tri-axial accelerometers, rate-gyroscopes, global positioning satellite systems and various vehicle diagnostic ports (including velocity and heading data) may be generated. This data may be automatically stored to built-in EDRs in case of an accident. The fitting of telematics units in vehicles is also well-known. Such telematics units generally comprise a mobile phone/cell phone transceiver to communicate information about the vehicle between the telematics unit and a remote processing entity.

Integrated navigation systems are also known. Such systems combine the outputs of several different types of sensors, for example by combining dead-reckoning calculations based on the output of an inertial measurement unit with position determination from a global positioning satellite system. As an example, systems integrating an inertial navigation system (INS) and a global navigation satellite system (GNSS) - known as INS/GNSS systems - are known.

Thus, crash forensics investigators these days often have available to them rich data for investigating crashes. However, a crash is usually a highly non-linear event. In general, vehicle motion during a crash is a motion in 3D space. Unlike under regular driving conditions, common non-holonomic constraints (such as limiting velocity in the lateral and vertical directions) are usually not applicable. Moreover, all the above-listed sources of data are error prone, especially in an extreme-dynamics environment. As a consequence, accuracy of forensic output is still very limited and traditional forensic tools can provide only lines of enquiry to investigators.

Thus, even with a high level of expertise and rich data, it is often difficult to resolve ambiguities. Complex and fatal road vehicle crashes therefore remain a matter of dispute and further improvement of crash forensics is consequently strongly desirable.

US patent 6,067,488 discloses a vehicle driving recorder and travel analyser in which an accelerometer and a gyroscope are provided for each of three dimensions. The recorder records up to 10 minutes worth of data from these sensors. If a shock (crash) occurs, the recorder continues to record this data for a predetermined amount of time and then stops. Subsequently, the travel analyser retrieves the data and performs zero- velocity drift correction of the gyroscope data based on the retrieved data. The analyser then determines the final travel attitude angles and reconstructs the vehicle's trajectory during the crash. However, the apparatus requires high cost, highly accurate sensors to achieve any degree of accuracy, and it is not clear even then that it would work as disclosed sufficiently well, or at all.

Summary of the Invention

The present invention is intended to address the shortcomings of the known art and has as an object to improve the accuracy of forensic analysis of crash events. In particular, the present invention is intended to provide an apparatus, system and method providing accurate travel data for calculating a vehicle three-dimensional trajectory before and after a crash. The apparatus may be self-contained on the vehicle or may be a distributed apparatus, for example comprising a telematics unit on the vehicle and a remote processing entity.

Elements of the present invention may be achieved by storing a set of vehicle travel data that enables objective analysis and reconstruction of the vehicle trajectory in a crash-proximity situation in order to elucidate accident causes. The present invention may combine different types of sensors and may use some specifics of vehicle crash situations to modify the existing use of integrated navigation systems in order to raise the accuracy of vehicle trajectory estimation.

According to a first aspect of the present invention, there is provided an apparatus for use in reconstructing a vehicle trajectory, comprising: a processing and control unit; and a memory, the apparatus being adapted to: capture inertial data at a predetermined frequency from an inertial unit mounted on the vehicle, the inertial unit including a 3D inertial sensor with 3D gyroscope functionality; repeatedly update a sensor error model and store the updated sensor error model; detect an event; and after the event is detected, store the captured inertial data until a predetermined time after the start of the event.

Preferably, the apparatus is adapted to receive external data and to update the sensor error model based on the external data.

Preferably, the apparatus is further adapted, each time the inertial data is captured, to determine whether the vehicle is in a steady state.

In this case it is preferred that the apparatus is further adapted, if the vehicle is in the steady state, to update the sensor error model based on the received inertial data using a zero-velocity update method.

It is also preferred that the apparatus is further adapted to compensate for errors in the received inertial data based on the updated sensor error model.

It yet more preferred that the apparatus is further adapted to calculate a vehicle state including at least one of an averaged steady acceleration vector for the period in which the vehicle has been in the steady state, a roll of the vehicle and a pitch of the vehicle based on the compensated inertial data.

Preferably, the apparatus is also further adapted, if the vehicle is not in a steady state, to calculate a predicted vehicle state based on a previous vehicle state and the inertial data. Preferably, the apparatus is adapted to receive external data and, if the external data is available, to update the sensor error model based on the inertial data and the external data.

In this case, it is further preferred that the apparatus is adapted to update the sensor error model based on a difference between a first vehicle state, which is determined using a previous vehicle state and the inertial data, and a second vehicle state, which is determined using the external data.

It is even more preferred that the second vehicle state is determined by updating a previous vehicle state using the external data.

Preferably, the external data in any of the foregoing comprises at least one of GPS data, GNSS data, other satellite tracking data, velocity-related data and temperature data.

The apparatus may comprise the inertial unit.

It is preferred that the predetermined time is determined as one of a fixed time after the start of the event and a minimum time throughout which a variation in the captured inertial data remains below a predetermined threshold.

Preferably, the apparatus is further adapted to store the captured inertial data before the event is detected.

Preferably, the captured inertial data is stored with a higher frequency after the event is detected.

In either case, it is preferred that the inertial data is captured with a higher frequency after the event is detected.

The event may be a crash. Preferably, the apparatus is adapted to update the sensor error model after the event has been detected based on inertial data stored after the event has been detected and when the vehicle is in a steady state.

Preferably, the apparatus is adapted to update the inertial data stored after the event is detected based on the most recently stored sensor error model, whereby the trajectory can be reconstructed based on the updated inertial data.

Preferably, the apparatus is adapted to determine resting position data of the vehicle after the event based on external data, whereby the trajectory can be reconstructed taking the determined end position as a starting point.

In this case, it is preferred that the resting position data includes at least one of data relating to the attitude of the vehicle and satellite positioning data.

Preferably, the apparatus is adapted to determine at least one of an averaged position, an averaged acceleration vector, and an averaged final heading using updated of data stored during a fixed period after detection of the event, whereby the trajectory can be reconstructed based on the determination.

In this case, it is preferred that the apparatus is further adapted to determine at least one of a final pitch, a final roll and a final yaw calculated using updated data stored during a fixed period after detection of the event, whereby the trajectory can be reconstructed based on the determination.

Preferably, the apparatus is adapted to reconstruct the trajectory by calculating at least one of a vehicle position, speed and attitude for each of a plurality of the first predetermined periods after the event using the updated stored sets of data.

Preferably, the apparatus is adapted to reconstruct the trajectory of the vehicle after the event based on the stored inertial data.

According to a further aspect of the present invention, there is provided a vehicle trajectory reconstruction apparatus comprising: a processing and control unit; and a memory, the apparatus being adapted to: capture data from an apparatus as described above; and reconstruct the trajectory of the vehicle based on the captured data.

According to a further aspect of the present invention, there is provided a system comprising both of the apparatuses described above.

According to a further aspect of the present invention, there is provided a method of obtaining data for use in reconstructing a vehicle trajectory, comprising: capturing inertial data at a predetermined frequency from an inertial unit mounted on the vehicle, the inertial unit including a 3D inertial sensor with 3D gyroscope functionality; repeatedly updating a sensor error model and storing the updated sensor error model; detecting an event; and after the event is detected, storing the captured inertial data until a

predetermined time after the start of the event.

Brief Description of the Drawings

Embodiments of the present invention will now be described by way of further example only and with reference to the accompanying drawings, in which:

Fig. 1 is a schematic drawing of a unit suitable for use in the present invention;

Fig. 2 is a schematic representation of a system suitable for use in the present invention;

Fig. 3 is a further schematic view of a unit suitable for use in the present invention;

Fig. 4 is a schematic view of a 3D inertial sensor with 3D gyroscope functionality;

Fig. 5 shows a timeline of a crash useful for explaining vehicle trajectory reconstruction;

Fig. 6 is a graphical representation of a principal direction of force (PDOF);

Fig. 7 is a flow diagram showing a crash detection algorithm;

Fig. 8 is a flow diagram showing a steady state detection algorithm;

Fig. 9 is a flow diagram showing travel data recording

Fig. 10 is a flow diagram showing error estimation and correction of the inertial sensor unit outputs;

Fig. 1 1 is a flow diagram showing the determination of vehicle trajectory;

and

Fig. 12 is a schematic summary of various possible elements of the present invention. Detailed Description

The present invention provides an apparatus, system and method for calculating three dimensional trajectory of the vehicle in near crash conditions.

A first embodiment of the present invention comprises a unit 1000, as shown in Fig. 1 , which may be installed in a vehicle (not shown). As shown in Fig. 1 , the unit 1000 is a telematics unit (shown as T-Box 1000) and has three parts: a central part 100, a 6 degrees of freedom inertial unit 200 and additional components 310-360. However, the telematics functionality is not an essential element of the present invention.

The unit 1000 is mounted within a vehicle by one or any number of possible mounting options. The unit 1000 may be installed in an after-market process within the vehicle (meaning after the complete vehicle as such is fully assembled), or may be integrated into the vehicle during assembly. The unit 1000 is connected to the vehicle DC power supply and can, but not need not, be connected to the vehicle controlling and processing system.

The central part 100 of the unit 1000 includes global positioning system receiver 1 10, long distance wireless transceiver 120 and processing & controlling unit 130. Global positioning system receiver 1 10, which is not required in all embodiments, receives satellite signals to calculate a position of the unit 1000 using a satellite system such as GPS, Galileo, GLONASS, COMPASS, QZSS and may have specific accuracy enhancement functions. The overall position of the unit 1000 may be derived from a combination of information from different satellite location systems. The receiver system 1 10 may be realized within the unit 1000 either by a module providing localization data (geographical coordinates) or by providing signals to the processing unit 130, which may calculate location data, besides other independent functions it undertakes. Global positioning receiver system 1 10 may be realized by a plurality of technologies and use an integrated antenna and/or an external antenna. This external antenna may be placed inside of the unit 1000 enclosure (outside of the global positioning receiver system module 1 10) or outside of the unit 1000 enclosure. In addition, other types of radio localisation receivers might be used, which may but need not be terrestrial-based systems.

Long distance wireless transceiver 120 has the function of receiving and transmitting data (including raw data, and /or audio signals and/or video signals), with or without compression and inherently imposed and optionally added additional encryption. Long distance wireless transceiver 120 is not essential where it is not required to operate the invention as a telematics unit. Long distance wireless transceiver 120 typically uses cellular (mobile communication network) connectivity by one or a combination of systems:

a) Generation 2 (2G) mobile communication System (GSM, GPRS) b) Generation 2. 5 (2. 5G) (EDGE)

c) Generation 3 (3G) (UMTS, WBCDMA, HDCPA)

d) Generation 4 (4G) (LTE)

and/or systems like WiMax, and/or satellite communication systems, and/or other data transfer radio systems.

The global positioning receiver system 1 10 and the long distance wireless transceiver 120 are not required in all embodiments of the invention and may optionally be realized and utilized in the unit 1000 as a single module.

Processing & controlling unit 130 is realized by any one of a plurality of known CPU solutions, whereby preferably a 32 bit processor optionally combined with DSP is preferred.

The CPU processor can use no or any operating system (OS), for example based on Linux, a Microsoft-based OS or another type of OS such as RTOS, VX Works and Android. An embedded Linux solution is preferred.

6 degrees of freedom inertial unit 200 is a 3D inertial sensor with 3D gyroscope functionality. It preferably comprises a 3D MEMS accelerometer 210 and a 3D MEMS gyroscope 220. 3D MEMS accelerometer 210 may be realized physically by using a single chip, more than one chip (typically one per direction per axis) or a module based on MEMS accelerator sensors. 3D MEMS gyroscope 220 may be realized physically by using a single chip, more than one chip or a module based on MEMS technology. The usage of devices realized by MEMS technology (Micro Electro-Mechanical Sensors) or NEMS (Nano Electro-Mechanical Sensors) enables the devices to be of small size and light-weight, and simplifies assembly of the proposed unit 1000 PCB assembly. The 3D MEMS acceierometer 210 and 3D MEMS gyroscope 220 may be provided as a single chip or a single module.

Components 310-360 are also optional components. Memory 310 may be realized using any suitable technology and can optionally be part of the memory of the processing and control unit 130. Preferably, where provided, the memory 310 comprises a non-volatile memory in which programming for the processing and controlling unit 130 and various coefficients are stored and a volatile memory, which may provide a working memory for the processing and controlling unit 130. The memory 310 provides resources for storing one or more of:

• data before transmission over long range wireless transceiver 120

• identification data of the vehicle

• access, maintenance, and service data

• business process relevant data

• driving event data records related to the vehicle in which the unit 1000 is mounted

• event data profiles required to detect and react to a specific event

• location based information with time stamps related to the vehicle

• driver behaviour data associated with the specific pre-defined events with time stamps or statistically evaluated without time stamps

• vehicle dynamic data (such as speed vectors and acceleration vectors) associated to the specific pre-defined events

Short range wireless connectivity 320 allows short range wireless data exchange between the unit 1000 and a remote unit, for example where the remote unit is less than 500 metres, and typically less than 20 metres, away from the unit 1000. It may be realized by any one of a plurality of well-known short range wireless solutions, such as one or more of:

• Bluetooth Systems in the 2. 4 GHz band • WLAN Systems in the 2. 4 & 5 GHz band

• ISM Band Systems in the 433 MHz, 866MHz, 315MHz, 915 MHz bands typically using protocols with limited duty cycles and typically 200kbit/s max raw data rate in communication

• UWB systems in the 3-10 GHz range

• 60 GHz - 24 GHz communication systems

• 24 GHz communication systems

• 60-80 GHz Radar Systems

• 24 GHz Radar Systems

Short range wireless connectivity 320 allows:

• wireless connectivity to an in-vehicle system; the unit 1000 may obtain internal information from the vehicle systems and use it for purposes such as event detection and related actions with dedicated time stamps

• wireless connectivity for additional sensors such as wireless camera connection, or driving environment sensors

• wireless connectivity to a driver's own independent personal information devices (PDA, Smart Phone or similar)

• providing sensory activity by itself for purposes of distance calculations or object recognition, by deploying external connectors for additional antenna systems.

Connections to or the provision of sensor(s) 330 allows wired means of connection to a specific non inertial sensor, being placed in the unit 1000 itself or outside of the unit 1000, for example sensors for environmental factors, odometers or magnetometers.

Wired interface to vehicle system and accessories 340 provides wired means for connection of the unit 1000 to vehicle systems or accessories by at least one of the means:

• Vehicle OBD Connector

• CAN Interface

• Lin Interface

• FlexRay Interface • MOST Interface

• SPI Interface

• RS232 Interface

• USB Interface

Microphone 350 is used for audio capture.

Speaker 360 contains can be used to issue alerts from the unit 1000 to the vehicle and the driver or to transmit alerts. A display (not shown) can also be used to provide the driver or another person with alerts and other information.

As shown in Fig. 2, the telematics unit 1000 can be connected to a remote processing entity 2000 or back end by means of a long distance wireless network 3000, typically a cellular or mobile phone network. These components together form a system 4000 of another embodiment of the invention, which will be discussed in more detail below. It should be noted that more than one telematics unit 1000 can be connected to the remote processing entity 2000.

As schematically represented in Fig. 3, the unit 1000 can receive a number of inputs and carry out a number of functions. In particular, the unit 1000 receives input data, provided to the processing and control unit 130 and the memory 310 in order to execute a variety of functions, the input data including:

• inertial unit data (such as acceleration and speed vectors) typically provided by the 3D inertial sensor with 3D gyroscope functionality 200; and

• location data from satellite positioning systems, typically provided by the global positioning receiver system 1 10; and/or

• data from the vehicle system to which the unit 1000 is mounted, typically provided by the wired interface 340; and/or

• data provided by the additional sensors (environment, accessories) 330.

The input data may also include:

• control data (settings, orders) typically provided by the back end 2000; and • maintenance and upgrade data typically provided by the back end 2000

Based on this received data, the processing and control unit 30 carries out a number of functions as discussed in more detail below. In addition to the operations described in more detail below, the process and control unit 130 may optionally carry out any one or more of:

• calculation of real time position data 1 1 100

• calculation of real time vector trajectory of the vehicle 1 1200

• calculation of behaviour of the driver & vehicle 1 1300

• calculation of event detection 1 1400

• calculation of vector trajectory of the vehicle after event occurrence 1 1500

• calculation of pre-event warning to vehicle system (driver) 1 1600

• encryption and multimedia compression 1 1700

• initialization of event related alerts 1 1800

As shown in Fig. 4, the inertial sensor 200 is able to detect acceleration 'a' as a vector having a magnitude in the direction of acceleration of the vehicle. In particular, the acceleration vector has a scalar component ax, ay, a_∑, in each of three orthogonal axes (X, Y, Z), which are measured by the inertial sensor 200. In addition, the inertial sensor 200 is able to detect angular acceleration about each of the axes, where <¾, <¾, αψ are the angular accelerations about the X, Y, Z axes respectively. Thus, using the inertial sensor 200, the unit 1000 is able to detect scalar acceleration information in predefined time periods as well as changes in the acceleration vector in the same periods. Moreover, given that the initial velocity will be known (zero before the vehicle moves), it is possible to calculate the velocity (both in terms of scalar velocity and velocity vector changes) and the roll, pitch and yaw rates, <¾, <¾, ωψ about the X, Y, Z axes respectively of the unit 1000, and hence of the vehicle. Moreover, the use of the inertial unit 200 having six degrees of freedom allows determination of the real time vector trajectory of the vehicle, as well as the position and attitude of the vehicle (including its degree of roll, pitch and yaw) at any one time.

Before providing a more detailed description of the operation of the unit 1000, it is worth discussing the drawbacks of existing apparatuses in more detail. Inertial navigation is a complex mathematical procedure that combines inertial sensor outputs to provide autonomous navigation of an object in space. This is usually called dead reckoning (DR). In particular, dead reckoning can be used to determine a vehicle's motion as the evolution of the position and orientation of the vehicle, as well as the time derivatives of these components, using equations of motion.

The conventional integration of the relative motions over time from a known position is challenging, particularly for low cost sensors such as MEMS and NEMS sensors because of the fast accumulation of errors. MEMS and NEMS sensors have extreme stochastic variances and error characteristics that change very quickly. In view of this, it is known to use a sensor error model to correct for sensor errors. A very simplified sensor error model under low-dynamic, normal driving conditions comprises only coordinate frame misalignments, and accelerometer/rate-gyroscope biases and scale factors.

Although dead reckoning is well-known, it is difficult to accurately determine a vehicle's motion during a crash using dead reckoning due to the complexity of the event. As noted above, crash situations are characterized by extreme dynamics and a crash is a complex and non-linear event. Moreover, during a crash a vehicle may not act as a rigid body due to the severe transformation of kinematic energy to structural deformation and a significant crush zone.

A crash pulse is the deceleration time history at a selected point in the vehicle during impact. A crash pulse amplitude during severe collisions may reach levels up to 50- 100g. In some parts of the vehicle, vibrations of several thousand g can be measured. This strongly depends on the vehicle structure, the measurement site and the impact mode.

Thus, although inertial sensors have been used for automatic road vehicle collision detection and post-crash analysis for decades, in the conditions involved in a crash, the performance of low cost sensor technologies such as MEMS and NEMS sensors is generally considered to be sub-optimal for dead reckoning and, particularly, vehicle trajectory reconstruction after a crash. In particular, their performance is severely degraded due to "g-sensitivity" (especially on rate-gyros), the non-linearity of accelerometers and the occurrence of angular acceleration sensitivity during vehicle rotation. The influence of high g-forces on sensors is commonly rated by manufacturers, with the complexity of these effects usually being a higher order function. Moreover, trajectory reconstruction accumulates sensor errors, so that the effect of sensor errors increases on a quadratic and cubic basis as the calculated trajectory progresses. Thus, the behaviour of low cost inertial measurement units (IMUs) brings a large degree of inaccuracy to the reconstruction of vehicle trajectory after a crash.

For measurements in extreme dynamics conditions such as occur during a crash, it is therefore common to use multiple sets of tri-axial accelerometers (such as a set of each of low, medium and high dynamic range accelerometers) to obtain a fused signal that combines the best properties of each: high range, low noise, good resolution and stability.

The inventors have recognised that three main sources of errors should be considered:

1 . Inertial sensor drifts or errors

2. Initialization - levelling and orientation errors and initial position and velocity errors

3. Environmental influences (such as temperature)

In addition, acceleration measured in the body frame (the vehicle's frame of reference using XYZ co-ordinates) will contain errors and there are errors in translation of measurements in the body frame to the navigation frame (north, east, down or NED) due to improper calculation of the vehicle platform orientation. These attitude errors additionally increase errors in calculations of rotations ^ω/^ζ (Earth rotation projections on NED-North-East-Down coordinate frame axis) and^ws.v (angular velocity of rotation of a navigation coordinate frame relative to the Earth). The difference between the true orientation of the navigation frame and the platform orientation is then further affected. In short, when using inertial sensors it is necessary to perform co-ordinate transformations between the frame of reference of the inertial sensors, the frame of reference in which the sensors are mounted and the frame of reference in which results should be presented, such as a geographic frame of reference. These transformations will each introduce errors. Further sources of error include Coriolis acceleration, and gravity and Earth-model imperfections. Accelerometer and rate gyro sensor imperfections are usually modelled with 7-8 different parameters and random variables. However, not all errors have the same impact on the navigation solution. In considering low-cost IMUs and short-term navigation (that is, the relatively short time over which dead reckoning must be performed to reconstruct post-crash vehicle trajectory), the inventors have recognised that the contribution of some types of errors is less significant and can be omitted from observation.

The largest positional drift with respect to time is due to residual gyroscope error. This source of error progresses by a cubic law with time and consumer grade MEMS gyroscope bias error typically induces position drift of a couple of dozens of meters after one minute. However, the inventors have recognised that, considering a unit with MEMS sensors and reasonable device installation accuracy (so the device is fitted level and with the correct orientation angles), the highest influence on short term position error during crash conditions will originate from an initial speed error. Accelerometer bias will be the second most influential cause of error. Gyro bias will be the third ranked error source. On a short run (the time required for post-crash vehicle trajectory reconstruction) bias more seriously contributes to position errors than sensor drift.

This analysis assumes that gyroscope bias error is fixed throughout the period of interest. In general, however, sensors have significant in-run bias instabilities. In other words, on a long run (for example, a journey during normal driving conditions) bias instabilities are large. Without frequent tracking of gyroscope bias changes, these errors would quickly drift and dominate the position error contribution.

In crash test labs, a commonly used instrumentation mounting point is the lowest body panel on the side of the car located between the two wheel wells at the pillar between front and rear doors, which is known as the B-pillar. It is presumed that the crash pulse at that point identifies the structural behaviour and the gross motion of the vehicle in a frontal impact. Preferably, fitting of the unit 1000 of the present invention should be compliant with this practice to ensure proper event severity characterization. In general, it is not necessary for various sensors to be collocated and their position usually does not match vehicle's centre of gravity. This means that there exists a nonzero lever-arm offset, so different sensors and the vehicle undergo slightly different motion patterns. To achieve higher trajectory reconstruction accuracy, a well-known lever-arm compensation technique may be used. Lever-arm compensation is usually carried out by mathematical translation of all sensors to one common reference point, by knowing exact distances between these points.

After the unit 1000 is mounted, on ignition (generally after every ignition and not only after the first one), a device initialization is carried out. Initialization comprises estimation of rate gyro drifts and horizontal levelling (calculating roll and pitch angles), but it can be also used for estimation of other error parameters, for example using feedback from coupling the inertial navigation system (INS) comprising the inertial unit 200 and the global navigation satellite system (GNSS) comprising the global positioning receiver system 1 10 in a steady position.

The invention proceeds on the basis that most single crash events can be described using the timeline shown in Fig. 5, in which:

(1 ) Pre-crash is a low/moderate dynamics interval. It is possible to use a dead- reckoning algorithm based on INS/GNSS integration for tracking sensor errors. This segment is continuous from the device initialization period. Regular vehicle dynamics are usually 2g or less.

(2) Crash is an interval with high system dynamics and unconstrained car motion as a reaction to extreme forces. The duration of the crash pulse is variable, but almost never exceeds 250 ms. There are many known algorithms for automatic crash detection. The angle of forces at the moment of impact relative to vehicle body frame is called the Principal Direction of Force (PDOF), which is described in more detail below.

(3) Post-crash is a low dynamics interval and typically lasts less than 10s, before vehicle gets to resting position.

(4) Resting position is an interval in which the vehicle is in a steady state after a crash. Accelerometer and gyroscope sampling rates in the present invention should preferably be above 100 Hz to enable accurate crash detection. GNSS usually has a limited dynamic range and typically can bring less value to the "crash" interval than during other segments. The interval following the crash pulse, identified as "post-crash", lasts as long as the vehicle is still moving. It has variable length. Its end, or the start of a steady state identified as the "resting position", is determined by means of measuring variance of output of the accelerometers. The description of this embodiment will focus on a single-crash situation. However, elucidation of multiple crashes may be deduced.

In view of the above, the unit 1000 is adapted to carry out a method of the present invention by running a number of algorithms, described in more detail below.

In general, however, the first step in reconstruction of the vehicle trajectory is performed on logs gathered while the vehicle was in a resting position. The method repeats parts of alignment procedures carried out in device initialization. Alignment may provide a coarse azimuth from a magnetic compass, and involves calibration of rate gyro drifts and horizontal levelling (determination of roll, pitch and optionally yaw angles of the vehicle in the resting position). Horizontal levelling is generally affected by sensor noise, vibrations and other sensor errors (biases). In one aspect of the present invention, deterministic parts of sensor biases are considered to be known from the pre- crash interval as a result of an INS/GNSS coupling algorithm and device initialization (or optionally only as a result of a continual update of errors in the INS). In the resting position, the outputs of sensors will not be affected by vehicle vibrations and some longer-term sensor statistics may be gathered to refine post-crash initialization/alignment. That is why the duration of resting position should preferably be more than 10s. Preferably, the duration should match the minimum of the graph of the square root of Allan variance.

Accordingly, by determining the corrected outputs of the sensors, and hence the corrected vector trajectory of the vehicle, scalar velocity information, scalar acceleration information, velocity vector changes and acceleration vector changes, it is possible accurately to reconstruct the vehicle trajectory in crash proximity in three-dimensional space. In the present embodiment, the processing and controlling unit 130 receives data from the inertial unit 200 and from "external" sources, such as the global positioning receiver system 1 10, sensors 330 and the vehicle 340 and runs a plurality of algorithms as discussed in more detail below to allow accurate trajectory reconstruction. In particular, in the present embodiment, the processing and controlling unit 130 receives data from:

• the inertial unit 200

• the global positioning receiver system 1 10 or GNSS, and

• the wired interface to vehicle systems and accessories 340.

In the discussion of the algorithms that follows, this data as a whole is referred to sensor data or sensor samples. Data from the inertial unit 200 is referred to as inertial data or inertial sensor samples. Data from the global positioning receiver system 1 10 and the wired interface to vehicle systems and accessories 340 is referred to as external data or external sensor samples. The unit 1000 runs the plurality of algorithms in parallel to detect crash events and store data required for vehicle crash trajectory reconstruction and, optionally, crash severity classification.

In the following description, "vehicle velocity" is defined as the moving velocity of the vehicle in the longitudinal or X direction. "Longitudinal acceleration" is defined as an acceleration component longitudinal to the direction of driving (that is, the component of acceleration in the X direction) during a specified time increment. Thus, if the vehicle is driving along the X-axis shown in Fig. 4, and the Z axis is vertical, the longitudinal acceleration will be defined as the acceleration in the X axis. Similarly, "lateral acceleration" is defined as an acceleration component perpendicular to the direction of driving (that is, the component of acceleration in the Y direction) during a specified time increment. Similarly "Yaw rate" is calculated as the "angular rate" or angular speed (ωψ) about the axis orthogonal to the vehicle plane - in other words, the Z-axis with the vehicle in the X-Y plane.

The "Principal Direction of Force (PDOF)" is defined as the angles of impact in the horizontal and vertical plane, as shown in Fig. 6. Thus, the PDOF comprises a horizontal PDOF and a vertical PDOF. The horizontal PDOF is the angle between the principle direction of force and the X axis in the X-Y plane, and the vertical PDOF is the angle between the principle direction of force and the X-Y plane. The PDOF is estimated in the sensor or vehicle "body" frame using simple trigonometry and signal filtering, as known to those skilled in the art. The PDOF value in the body frame can be translated to a value in the navigation frame (or absolute frame of reference) and depends on the vehicle's attitude at the moment of collision. Similarly "final roll" and "final pitch" angles are defined as the angles of the vehicle in the horizontal plane relative to X-Y and Y-Z planes respectively.

In more detail, following initialisation (preferably on ignition of the vehicle engine), the unit 1000 carries out travel data recording in accordance with the travel data recording algorithm shown in Fig. 9, which includes subsidiary algorithms for crash detection (Fig. 7), steady state detection (Fig. 8), sensor error model estimation (Fig. 10) and crash trajectory reconstruction (Fig. 1 1 ). However, these algorithms may be distributed across different units/devices. In particular, any one of the algorithms may be carried out by a remote processing entity. It is particularly envisaged that the crash trajectory reconstruction may be carried out by a remote unit having first obtained relevant data from the vehicle mounted unit 1000 either via the long distance wireless transceiver 120 or the short range wireless connectivity 320, or even by a wired connection made by crash investigators post-crash.

As noted above, it is preferred to use a sensor sampling rate of 100 Hz or greater and hence the travel data recording algorithm and hence the subsidiary algorithms (except the crash trajectory reconstruction algorithm) are typically also carried out with this frequency.

Turning to the crash detection algorithm shown in Fig. 7, first a value for an observation time window "observation window dV", which will typically be set between 20 and 50 ms is retrieved from the memory 310. "CRASH DETECTED" is a Boolean variable, and its initial state is FALSE or O.

The algorithm starts in step S100 by reading accelerometer data. In step S1 10, the algorithm calculates "delta velocity" as the change of the overall vehicle velocity (magnitude of the velocity vector) by accumulating the acceleration vector over the "observation window dV" time. After this, the process moves to S120 where the current value of PDOF is determined. The unit 1000 then determines a value for a threshold "delta velocity threshold", which will be typically set using a predefined profile or a look-up table dependable on PDOF angles, usually in both the horizontal and vertical planes. In case of a full frontal crash (PDOF = {0 deg, 0 deg}), the matching threshold value would typically be in the order of 4g.

The algorithm checks if the crash detection condition is satisfied in step S130 by determining if the absolute value of "delta velocity" is larger than "delta velocity threshold". If this condition is met, meaning that a crash event has started, "CRASH DETECTED" is then set to TRUE or 1 in step S140, and value of "PDOF" is stored for later use or sent via wireless. Optionally, the algorithm can use the calculated variables ("delta velocity" and PDOF) for classification of the crash event in terms of severity in step S150. The algorithm then returns and waits for the next measurement at step S101 .

If the condition in step S130 is not met the algorithm returns and waits for next measurement at step S101 .

The steady state detection algorithm is shown in Fig. 8. First, a number of predetermined values are retrieved from the memory 310. These are a value for an observation time window "observation window 1 ", which will typically be set between 0.1 and 1 s; a value for a threshold "STEADY THRESHOLD", the value of which depends on sensor characteristics but is typically below 0.2 ms^"2. Optionally, a value for a threshold "SPEED THRESHOLD" can be pre-set as an additional or alternative condition for the detection of a steady state. If set, the value of SPEED THRESHOLD is typically below 2 ms^"1. "STEADY COUNTER" is a variable used to measure steady state duration (time), and its initial value is 0.

The algorithm starts in step S200 by reading accelerometer data. In step S210, the algorithm calculates "acceleration variance" as the variance of the overall acceleration vector over predefined "observation window 1 ". Alternatively, it would be possible to monitor the variance of the worst (most varying) of the XYZ or NED components. There are various standard ways to calculate variance, including absolute deviation, squared deviation, standard deviation etc and any suitable measure of variance may be used. For example, "acceleration variance" may be determined simply as the difference between the maximum and minimum values for "acceleration" held in the buffer at the current time. Strictly speaking, the result of a maximum - minimum calculation would not usually be considered as variance directly, but rather +/- 3 sigma interval where variance could be extracted as sigma², assuming that typically noise can be characterized as dominantly White Gaussian noise. A maximum - minimum or similar calculation falls within the definition of variance in the present specification.

After this, the process moves to S220 where the algorithm checks if the steady state condition is satisfied by determining if the absolute value of "acceleration variance" is below "STEADY THRESHOLD". Optionally, another condition may be introduced as well or instead - whether "vehicle velocity" is below "SPEED THRESHOLD". If the first condition is met (or both conditions are met), meaning that a steady state has started, the value of "STEADY COUNTER" is incremented by 1 in step S230 and then the process continues to S201 . In S201 the algorithm returns and waits for the next measurement.

If the condition in step S220 is not met (or one of the conditions is not met), which means that the unit is not, or is no longer, in a steady state, the value of "STEADY COUNTER" is reset to zero and the algorithm returns and waits for next the measurement at step S201 .

The travel data recording algorithm is shown in Fig. 9. First, a value for a threshold "STEADY INTERVAL", which will typically be set to be greater than 5 s, is retrieved from the memory 310.

"STEADY COUNTER" is a variable used to measure steady state duration (time), its value originating and being updated in the steady state detection algorithm. Each increment corresponds to the frequency at which the algorithm is operated. Thus, if data is sampled at a frequency of 100 Hz and the count is at 100, then 1 second has elapsed. "Circular buffer" is a buffer of a predefined length typically sufficient to store more than 10 seconds of sensor samples. It is continually updated with the sensor samples. In particular, the sensor samples are sequentially stored in the circular buffer, and when an amount of data exceeds the storage capacity, the oldest data is erased in sequence so as to update the circular buffer. Thus, at least the sensor samples received in the last 10 seconds will always be stored in the circular buffer.

"CRASH DATA SET" is a variable capable of storing a time-stamped array of sensor samples-type of data.

The algorithm starts in step S400 by reading new sensor samples. In step S410, the previously described steady state detection algorithm is performed with new accelerometer data as required input. Its result is an updated value of STEADY COUNTER. In the following step S420, the previously described CRASH DETECTION ALGORITHM is performed with the new accelerometer data as the input to update the value of "CRASH DETECTED" and, if appropriate, store the PDOF.

After this, the process moves to S430 where the algorithm checks if the condition for crash detection is satisfied by determining if the value of "CRASH DETECTED" is set to 1 or TRUE.

If the condition in step S430 is not met, which means that no crash is detected, the sensor samples are stored in the circular buffer in step S500. In the next step S510, the content of the circular buffer is preserved as "CRASH DATA SET". The process then continues to step S520 where an algorithm for estimating and updating the sensor error model, which is described in more detail below, is executed. The algorithm then returns and waits for the next measurement at step S401 .

If the condition in step S430 is met, meaning that a crash state has already started, new sensor samples are appended to "CRASH DATA SET" in step S440 and then the process continues to step S450. In step S450 the algorithm checks if the steady state condition is satisfied by determining if the value of "STEADY COUNTER" is above the threshold "STEADY INTERVAL". If this condition is not met, which means that time T3 at the end of the resting position period has not been reached yet, the algorithm returns and waits for next measurement at step S401 . By contrast, if this condition is met, meaning that time T3 has been reached and the post-crash steady state has started, a crash trajectory reconstruction algorithm, which is described in more detail below, is executed in step S460. Following this step, in S470, the value of the "CRASH DETECTED" indicator will be reset to 0 and then the process continues to S401 . In S401 the algorithm returns and waits for the next measurement.

In the estimating sensor error model algorithm, the "SENSOR ERROR MODEL" is defined as a set of offsets, drifts, temperature dependencies, nonlinearities, and statistical models that model errors of "sensor samples". Similarly, "vehicle state" and "predicted vehicle state" are defined as an estimate of the position, velocity vector and optionally the attitude of the vehicle.

In more detail, the estimating sensor error model algorithm is shown in Fig. 10. First, the value of the STEADY COUNTER described above, which is a variable used to measure steady state duration (time), is retrieved from the memory 310. Its value originates and is updated in the steady state detection algorithm. In addition, the value of a threshold "STEADY PERIOD" is retrieved from the memory. The value of STEADY PERIOD may, but need not, be the same as that of STEADY INTERVAL, described above.

The algorithm starts in step S300 by capturing new sensor samples data previously stored in the circular buffer, preserved values of the variables "vehicle state" and "predicted vehicle state", and the value of STEADY COUNTER. "Vehicle state" and "predicted vehicle state" are each a set of predetermined variables describing the state of the vehicle. Preferably, these variables include or describe at least the position and velocity of the vehicle. They may also include or describe the attitude (angles) of the vehicle (roll, pitch and/or yaw), as well as its acceleration, rotational velocity and rotational acceleration. In step S310, the algorithm checks if the steady state condition is satisfied by determining if value of STEADY COUNTER is larger than the threshold STEADY PERIOD. If this condition is met, meaning that a steady state suitable for update of the sensor error model based on the INS has started, the algorithm continues to step S31 1 . In step S31 1 "SENSOR ERROR MODEL" is updated using known algorithms, typically known as a zero-velocity update technique. Such techniques may be based at least in part, for example, on the assumption that accelerometers do not accumulate drift, so the attitude angles represented by the acceleration data from accelerometers when the vehicle is stopped are accurate. By contrast, angular speedometers such as gyroscopes do accumulate drift. Accordingly, the difference between attitude angles derived from angular speedometers and accelerometers in the steady state is an indication of drift. In this way, the drift tendencies of gyroscopes and offset values can be corrected. The updated SENSOR ERROR MODEL is then stored for future use.

In the next step, S312, which is introduced for greater clarity, "predicted vehicle state" will keep its old value. In an alternative embodiment, if "vehicle state" and "predicted vehicle state" also contain attitude (angles) information it is possible to update the roll and pitch angles of "vehicle state" and "predicted vehicle state" based on the steady state, by means known to persons skilled in art.

In the following step, S340, the process checks if new "external sensor samples" are available to the algorithm and, if so, the process continues to step S350.

In step S350, the "vehicle state" is updated according to new external measurements given in an "external sensor samples" data set. The external data may include, for example, any one or more of the position of the vehicle determined by the GNSS (global positioning receiver system 1 10), temperature data from external sensor 330, and speed and/or acceleration data from the wired interface to the vehicle system 340. In particular, the externally measured speed may be substituted for the vehicle velocity component of the current "vehicle state", and the externally derived position may be substituted for the position component of the current "vehicle state".

Subsequently, in step S360 the difference between the (one or more of the variables in) "predicted vehicle state" and the (corresponding variables in) updated "vehicle state" is calculated. This difference is termed "innovation".

Next, in step S370, the "innovation" variable(s) is/are used to update the SENSOR ERROR MODEL. In particular, the "innovation" variable(s) is/are used to update the parameters of the sensor error model that correct for sensor biases, scale factors, gyro scale factors, gyro biases, cumulative drifts and so forth. In particular, a linear or nonlinear estimator (which may include any of Kalman filters, extended Kalman filters (EKFs), particle filters, and unscended Kalman filters) is used to update the SENSOR ERROR MODEL based on the "innovation" variable(s).

The process continues to step S380, where "predicted vehicle state" is updated to values contained in "vehicle state", for future use.

In the next step S390, the most recently updated SENSOR ERROR MODEL, "vehicle state" and "predicted vehicle state" as intermediate and auxiliary results are stored to memory.

The algorithm then returns and waits for the next measurement at step S301

If the condition in step S310 is not met, which means that no steady state suitable for sensor update by the zero-velocity method is detected, the newly read "inertial sensor samples" are compensated for errors using the existing SENSOR ERROR MODEL in step S320 and for clarity are termed "corrected inertial sensor samples".

The process continues to step S330 where the "predicted vehicle state" is calculated using the "corrected inertial sensor samples" and the previously stored "predicted vehicle state" by means of solving navigation equations to calculate a new position, velocity and attitude of the vehicle. After this, the algorithm continues to step S340, which is already described.

If condition in step S340 is not met, the algorithm continues to step S390, which has already been described.

The crash trajectory reconstruction algorithm is shown in Fig. 1 1 . This algorithm in general can have specific conditions and branches that cover particular cases such as multiple crashes, discontinued resting period, rapid change of environmental

temperature, a crush zone that introduces a need for more manual and post-event adjustments or similar. The basic form presented here will list steps as one sequence, with various optional steps. First, a value for threshold "STEADY INTERVAL", which is the same as the value preset in the travel data recording algorithm, is retrieved from the memory 310.

The algorithm starts in step S600 by reading values of "CRASH DATA SET" and "SENSOR ERROR MODEL", which were updated in the travel data recording algorithm and the sensor error model estimating algorithm respectively.

In step S610, the part of the "CRASH DATA SET" array that chronologically matches the "resting position" of the vehicle is copied to a "RESTING POSITION DATA SET" variable. In particular, the data stored in the "RESTING POSITION DATA SET" corresponds to the data that was appended to the "CRASH DATA SET" from the time the STEADY COUNTER was first updated to 1 until the time the STEADY COUNTER reached the threshold STEADY INTERVAL.

In step S620, which is optional, gyroscope biases as part of the "SENSOR ERROR MODEL" are calculated by averaging rate gyro outputs gathered in the "RESTING POSITION DATA SET". In step S630, which is also optional, the "SENSOR ERROR MODEL" is updated using the new information from S620.

In step S640, the "CRASH DATA SET" and the "RESTING POSITION DATA SET" are then compensated for errors estimated in the "SENSOR ERROR MODEL" (or the updated SENSOR ERROR MODEL if steps S620 and S630 have been carried out). The compensated data is stored as the new variables "CORRECTED CRASH DATA SET" and "CORRECTED RESTING POSITION DATA SET". It should be noted that "CORRECTED CRASH DATA SET" covers historical data of all sensor samples from the beginning of the "pre-crash" period to the end of the "resting position" period shown in Fig. 12. The skilled addressee will recognise that rather than extracting the "RESTING POSITION DATA SET" and then correcting both it and the "CRASH DATA SET", it is possible to correct the "CRASH DATA SET" and then extract the "CORRECTED RESTING POSITION DATA SET".

Next, in step S650 an "averaged steady acceleration vector" is calculated by averaging the updated acceleration values stored in the "CORRECTED RESTING POSITION DATA SET". This information is used in step S660 for the calculation of "final pitch" and "final roll" angles. Optionally, the "final yaw" angle may also be calculated.

The next step, step S670, is also optional. In it, external data relating to the resting position of vehicle can be further used to determine the "final heading" (corresponding to the yaw). This can be determined as the vehicle azimuth in the "resting position" by means of processing magnetometer data if available or updating the algorithm by expert witness findings. Similarly, the "final location" of the vehicle after the crash, usually defined via longitude and latitude, can be determined by means of GNSS, map matching, radio-triangulation or by expert witness data.

Next, step S680 defines a "FINAL VEHICLE STATE" variable using all the newly available data: "final roll", "final pitch", "final heading", "final location" and "final speed", which is typically zero, for easier further handling.

In step S690, a dynamic part of the "CORRECTED CRASH DATA SET" is copied in inverse order to a "REVERSE CRASH LOG" array. The dynamic part excludes data samples appended after the STEADY COUNT has started (that is, excludes data samples appended in the "resting position" period).

The next step, step S700, is a trajectory reconstruction step. In this step, the movement of the vehicle back from the defined "FINAL VEHICLE STATE" is determined based on the "REVERSE CRASH LOG" using known navigation equations and known kinematic methods that describe objects in 3D space. In particular, the vehicle state at each sampled point can be determined by solving navigation equations of a strap-down inertial navigation system and acceleration vector integration. For example, known methods use direction cosines, Euler angles, quaternions, dual quaternions and/or axial vectors. The calculated trajectory in step S700 is stored as "REVERSE VEHICLE 3D CRASH TRAJECTORY" as it is inverse to the natural crash timeline shown in Fig. 12.

In step S710, the "REVERSE VEHICLE 3D CRASH TRAJECTORY" is inverted back to the natural timeline and stored as "3D VEHICLE CRASH TRAJECTORY" and then the process continues to step S701 . In step S701 the algorithm returns and waits for the next crash detection. The present embodiment has a number of advantages over known apparatuses and methods of vehicle trajectory reconstruction. First, the present invention uses a sensor error model which is continuously updated during normal travel of the vehicle (that is, before a crash occurs). Because the sensor error model is continuously recorded up until a crash is detected (it is updated with a 100 Hz frequency in the present embodiment), it takes into account the whole history of the vehicle from the last initialisation up to detection of the crash. Hence, the data recorded up to and during the crash is accurate, or can be accurately reconstructed. In particular, the most recently stored sensor error model provides an accurate means of subsequently updating or correcting the recorded crash data set, which is the data recorded from the detection of a crash onwards.

By contrast, previous trajectory reconstruction has relied on carrying out the known zero-velocity update correction after a crash based on pre-crash samples stored, for example, in only the 10 minutes preceding a crash to determine the drift of the gyroscopes (angular speedometers). Thus, the previous trajectory reconstruction is considerably less accurate than the trajectory reconstruction of the present invention. It also requires a much larger memory to store 10 minutes worth of data as opposed to 10 seconds worth, or alternatively it requires a much lower sampling rate, further decreasing the accuracy.

Although not essential to the present invention, in the present embodiment, the processing and controlling unit 130 receives data from both the inertial unit 200 and from "external" sources, such as the global positioning receiver system 1 10, sensors 330 and the vehicle 340. The sensor error model estimation algorithm updates the sensor error model using a zero-velocity correction method if the vehicle is in a sufficiently steady state, for example because the variance in vehicle acceleration is below a predetermined threshold and/or the vehicle velocity is below a predetermined threshold. However, if these conditions are not met, the sensor error model algorithm causes the processing and controlling unit 130 to update the vehicle state and the sensor error model based only on the external sensor samples, if available. Given that it will commonly not be possible to update the sensor error model using a zero-velocity correction method in the period before a crash, since in most cases the vehicle will have been moving out of a steady state condition, updating of the sensor error model based on external samples at regular intervals allows a significantly improved sensor error model to be stored at all times.

In addition, because an accurate sensor error model updated before the crash is stored in the present embodiment, after the vehicle has come to a rest following a crash, this stored sensor error model can be used in step S640 to correct the crash data set stored in sequential iterations of step S440, thereby further improving the accuracy of trajectory reconstruction over the known art.

Furthermore, the calculation of gyroscope biases and/or gyroscope drift from the resting position data set and the use of the results to update the stored sensor error model before using it to correct the crash data set further improve the accuracy of trajectory reconstruction over the known art.

The use of external data to improve the definition of the final vehicle state (for example, the final roll, pitch, heading, location and speed) also improves the accuracy of the final trajectory reconstruction.

Accordingly, it will be appreciated that many separate and individual aspects of the present embodiment attain the goal improving the accuracy of the analysis of crash events. Together, these aspects provide a remarkable improvement in the reconstruction of vehicle trajectory after a crash event. One or more of these aspects allows accurate trajectory reconstruction using only a single set of low cost sensors, such as MEMS or NEMS sensors. Accordingly, the present invention makes the use of units such as the unit 1000 in the present embodiment a realistic prospect in everyday vehicles on public roads.

In the present embodiment, the SENSOR ERROR MODEL is updated repeatedly, and uncorrected sensor data is stored in the circular buffer and CRASH DATA SET. However, it would be possible to correct the sensor data using the most recently updated SENSOR ERROR MODEL and subsequently only use the SENSOR ERROR MODEL to correct previously uncorrected sensor data during trajectory reconstruction. In the present embodiment, in each cycle of the Estimating SENSOR ERROR MODEL algorithm, it is determined both whether the inertial unit 200 is in a steady state and whether external sensor samples are available. If the answer is yes in both cases, the SENSOR ERROR MODEL is updated using both the inertial data and the external data. However, this is not essential.

For example, it would instead be possible to determine first whether the inertial unit 200 were in a steady state and, if it were, update the SENSOR ERROR MODEL based only on the steady state inertial data. Alternatively, if the inertial unit 200 were not in the steady state, the SENSOR ERROR MODEL would be updated based only on the external data (if available).

Alternatively, it would be possible to determine first whether external sensor samples were available and, if they were, update the SENSOR ERROR MODEL based only on the external sensor samples. Alternatively, if the external sensor samples were not available, the SENSOR ERROR MODEL would be updated based only on the steady state inertial data (if available).

In the present embodiment, it is first determined whether the vehicle is in the steady state and then whether external sensor samples are available. However, these determinations could be made in the reverse order. The SENSOR ERROR MODEL updates could also be made in the reverse order, although this would not be essential.

Moreover, it would be possible to update the SENSOR ERROR MODEL based on only the steady state inertial data, or only on external sensor samples, without checking to see whether the other type of update is possible.

The present invention has been described with respect to an inertial unit 200 having a single set of low cost acceleration sensors and a single set of low cost gyroscopes, both based on MEMS technology. However, to further improve the accuracy of vehicle reconstruction, higher cost, more accurate sensors may be used. In addition, two, three or more sets of accelerometers and/or two, three or more sets of gyroscopes may be provided to gain accuracy. In this case the two or more sets will require time alignment and level alignment operations, and their output will need to be unified. In the case of sensor saturation, accelerometer or gyroscope output may be partially reconstructed using spline or other interpretation.

An alternative embodiment of the present invention is shown in Fig. 12. In particular, as shown at 1 120 the unit 1000 records to the circular buffer pre-crash accelerometer, gyroscope, GPS, odometer, magnetometer and other sensor data for a pre-defined period (for example, 10 seconds). In addition, as shown at 1 121 the unit 1000 calculates accelerometer, gyroscope, GPS, odometer, magnetometer and other sensor data error compensation matrices to correct for factors such as bias, scale factors, axis misalignment, temperature, noise, random walk, drift and so forth.

As shown in 1 1250, the unit 1000 also (or instead) carries out continuous tracking of accelerometer, gyroscope, GPS, magnetometer and other sensors and models them using linear and non-linear estimators, such as Kalman filtering, extended Kalman filtering, particle filtering and others. As shown at 1 1250 and 1 1251 , the results are used to estimate pre-crash vehicle angles (eg roll, pitch and heading/yaw) and a pre- crash velocity vector.

The unit 1000 detects a crash event as shown at 1 1380, and may classify crash events, for example into severe and non-severe crashes.

As shown at 1 1230, after detecting a crash (or optionally only a severe crash) the unit 1000 records to a buffer crash data including accelerometer, gyroscope, GPS, odometer, magnetometer and other sensor data for a pre-defined period or until detection of the resting position, which is shown at 1 1210.

Following that, as shown at 1 1240, the unit 1000 records resting position data including accelerometer, gyroscope, GPS, odometer, magnetometer and other sensor data for a pre-defined period throughout which the vehicle is in the resting position. In addition, as shown at 1 1241 the unit 1000 calculates accelerometer, gyroscope, GPS, odometer, magnetometer and other sensor data error compensation matrices to correct for factors such as bias, scale factors, axis misalignment, temperature, noise, random walk, drift and so forth. As shown at 1 1242, the unit 1000 also calculates final position angles from the resting data and the error compensation matrices. Finally, and optionally, as shown at 1 1260, the unit 1000 carries out correction of recorded data and/or estimated models using data from external sensors or measurements, for example post-crash inclinometer data 1 1261 , measurements carried out by experts (orientation, inclination, position) 1 1262 and temperature sensor data and compensation of temperature effects on inertial data 1 1263.

The foregoing is then used to calculate an accurate post-crash vehicle trajectory, in the same way as the first embodiment.

The skilled addressee will recognise that the various elements of the first embodiment and this embodiment are interchangeable. This embodiment provides some or all of the advantages of the first embodiment, as well as additional advantages.

Moreover, in order to carry out vehicle trajectory reconstruction, the telematics elements of the unit 1000 are not required in the present invention and can therefore be removed if desired. In particular, only the inertial sensor 200, the processing and control unit 130, and the memory 310 are generally essential and any two or more of these can be integrated. The location data 1 10 is also strongly preferable.

Alternatively, the unit 1000 can make use of other inputs to improve or otherwise adjust calculation of the vehicle crash trajectory. For example, the unit 1000 may make use of other types of positioning receiver system to place the vehicle on a pre-stored map or a map downloaded from the remote processing entity 2000 or another source via the long distance wireless transceiver 120, the short range wireless connectivity 320 or the wired interface 340.

Although the unit 1000 has been described as detecting crash events and calculating the vehicle crash trajectory, instead it may send either the data needed to determine whether an event has occurred or the sensor samples of the event (or both) to the back end 2000, which then carries out event detection and/or crash trajectory calculation for the individual vehicle. The processed data may be sent back to the unit 1000, or another device, whether or not on the vehicle. The connection to or provision of sensors 330 may allow the input of environmental conditions such as temperature that can be used for better sensor error characterization. As with the global positioning data, the input from the sensors can be used to modify the thresholds used in crash detection, resting position calculation and/or the interval durations. For example, if a crash is detected and the vehicle is in the resting position in strong winds and/or rain, this may indicate that the value of the threshold for steady state detection should be appropriately adjusted or the resting position should by detected by other means such as the odometer, whether the engine is running and/or on-board diagnostic data.

The detected crash events (optionally of different severities) and the reconstructed vehicle trajectory calculated by the present invention may be advantageously be used by vehicle insurance companies to estimate the cause of accidents and the risk posed by individual drivers, the collective risk on a vehicle driven by named drivers, and the risk on a fleet of vehicles as well as individual vehicles and drivers within the fleet. In addition, these factors can be used by insurers and fleet owners and operators to establish which vehicles or makes of vehicles are safest to drive, as well as which routes are safest, and which drivers should be singled out for additional safety training or in worst cases disciplinary action.

The foregoing description has been given by way of example only and it will be appreciated by a person skilled in the art that modifications can be made without departing from the scope of the present invention.

Claims

1 . An apparatus for use in reconstructing a vehicle trajectory, comprising:

a processing and control unit; and

a memory,

the apparatus being adapted to:

capture inertial data at a predetermined frequency from an inertial unit mounted on the vehicle, the inertial unit including a 3D inertial sensor with 3D gyroscope functionality;

repeatedly update a sensor error model and store the updated sensor error model;

detect an event; and

after the event is detected, store the captured inertial data until a predetermined time after the start of the event.

2. An apparatus according to claim 1 , adapted to receive external data and to update the sensor error model based on the external data.

3. An apparatus according to claim 1 or claim 2, further adapted, each time the inertial data is captured, to determine whether the vehicle is in a steady state.

4. An apparatus according to claim 3, further adapted, if the vehicle is in the steady state, to update the sensor error model based on the received inertial data using a zero- velocity update method.

5. An apparatus according to claim 4, further adapted to compensate for errors in the received inertial data based on the updated sensor error model.

6. An apparatus according to claim 5, further adapted to calculate a vehicle state including at least one of an averaged steady acceleration vector for the period in which the vehicle has been in the steady state, a roll of the vehicle and a pitch of the vehicle based on the compensated inertial data.

7. An apparatus according to any one of claims 3 to 6, further adapted, if the vehicle is not in a steady state, to calculate a predicted vehicle state based on a previous vehicle state and the inertial data.

8. An apparatus according to any one of claims 2 to 7, adapted to receive external data and, if the external data is available, to update the sensor error model based on the inertial data and the external data.

9. An apparatus according to 8, further adapted to update the sensor error model based on a difference between a first vehicle state, which is determined using a previous vehicle state and the inertial data, and a second vehicle state, which is determined using the external data.

10. An apparatus according to claim 9, wherein the second vehicle state is determined by updating a previous vehicle state using the external data.

1 1 . An apparatus according to any one of claims 2 to 10, wherein the external data comprises at least one of GPS data, GNSS data, other satellite tracking data, velocity- related data and temperature data.

12. An apparatus according to any one of the preceding claims, further comprising the inertial unit.

13. An apparatus according to any one of the preceding claims, wherein the

predetermined time is determined as one of a fixed time after the start of the event and a minimum time throughout which a variation in the captured inertial data remains below a predetermined threshold.

14. An apparatus according to any one of the preceding claims, further adapted to store the captured inertial data before the event is detected.

15. An apparatus according to claim 14, wherein the captured inertial data is stored with a higher frequency after the event is detected.

16. An apparatus according to claim 14 or claim 15, wherein the inertial data is captured with a higher frequency after the event is detected.

17. An apparatus according to any one of the preceding claims, wherein the event is a crash.

18. An apparatus according to any one of the preceding claims, further adapted to updated the sensor error model after the event has been detected based on inertial data stored after the event has been detected and when the vehicle is in a steady state.

19. An apparatus according to any one of the preceding claims, further adapted to update the inertial data stored after the event is detected based on the most recently stored sensor error model, whereby the trajectory can be reconstructed based on the updated inertial data.

20. An apparatus according to any one of the preceding claims, further adapted to determine resting position data of the vehicle after the event based on external data, whereby the trajectory can be reconstructed taking the determined end position as a starting point.

21 . An apparatus according to claim 20, wherein the resting position data includes at least one of data relating to the attitude of the vehicle and satellite positioning data.

22. An apparatus according to any one of the preceding claims, further adapted to determine at least one of an averaged position, an averaged acceleration vector, and an averaged final heading using updated of data stored during a fixed period after detection of the event, whereby the trajectory can be reconstructed based on the determination.

23. An apparatus according to claim 22, further adapted to determine at least one of a final pitch, a final roll and a final yaw calculated using updated data stored during a fixed period after detection of the event, whereby the trajectory can be reconstructed based on the determination.

24. An apparatus according to any one of claims 18 to 23, further adapted to reconstruct the trajectory by calculating at least one of a vehicle position, speed and attitude for each of a plurality of the first predeternnined periods after the event using the updated stored sets of data.

25. An apparatus according to any one of claims 1 to 24, further adapted to reconstruct the trajectory of the vehicle after the event based on the stored inertial data.

26. A vehicle trajectory reconstruction apparatus comprising:

a processing and control unit; and

a memory,

the apparatus being adapted to:

capture data from an apparatus according to any one of claims 1 to 24; and reconstruct the trajectory of the vehicle based on the captured data.

26. A system comprising an apparatus according to any one of claims 1 to 24 and an apparatus according to claim 26.

27. A method of obtaining data for use in reconstructing a vehicle trajectory, comprising:

capturing inertial data at a predetermined frequency from an inertial unit mounted on the vehicle, the inertial unit including a 3D inertial sensor with 3D gyroscope functionality;

repeatedly updating a sensor error model and storing the updated sensor error model;

detecting an event; and

after the event is detected, storing the captured inertial data until a

predetermined time after the start of the event.