SE544424C2 - Safety wearable for egress detection from a work vehicle - Google Patents

Abstract

A safety wearable (32) is configured to be worn by an operator of a work vehicle, and comprises data processing equipment configured to feed sensor data from a sensor arrangement as a feature vector to a trained classifier trained to detect a cabin egress event indicating that the operator exits or has exited an operator cabin of the work vehicle; and generate, based on a classification received from the trained classifier, a cabin egress event signal. The cabin egress event signal may e.g. enable a safety light (40).

Description

SAFETY WEARABLE FOR EGRESS DETECTION FROM A WORKVEHICLE Field of the invention The present invention relates to a safety wearable, a computer-implementedmethod of detecting egress from a work vehicle, data processing equipment and acomputer program product implementing such a method, and a computer-readablestorage medium having stored thereupon such a computer program product.

BackgroundWork vehicles are often used in complex environments - in the case of transportation trucks, in intense traffic and on busy loading and unloading sites, andin the case of e.g. construction machines, in environments where there may not evenbe any predefined roads or traffic rules. Obviously, working in such environmentsmay pose a danger to the operators of work vehicles.

Summafllt is an object of the present invention to solve, or at least mitigate, parts or all of the above mentioned problems. To this end, according to a first aspect, there isprovided a safety wearable configured to be worn by an operator of a work vehicle,the safety wearable comprising a sensor arrangement, and data processingequipment configured to: feed sensor data from the sensor arrangement as a featurevector to a trained classifier trained to detect a cabin egress event, the cabin egressevent indicating that the operator exits or has exited an operator cabin of the workvehicle; receive a classification from the trained classifier; and generate, based onthe classification received from the trained classifier, a cabin egress event signal inresponse to the classification indicating an egress event. Thereby, an alert mayconveniently and automatically be issued in response to an egress event, to warne.g. nearby and/or remote persons and/or machinery that the operator has exited thework vehicle. This increases the safety of e.g. a construction site, or during roadtransport. The work vehicle may be, for example, a cargo truck or lorry, a tractor, alifting truck such as a forklift truck, or a construction machine such as an excavator,an articulated hauler, or a wheel loader. The work vehicle may propel itself on wheelsor continuous tracks.According to embodiments, the trained classifier may be implemented in said data processing equipment. Providing the trained classifier within the data processingequipment comprised in the safety wearable enables fast and reliable egressdetection. Feeding the sensor data to the trained classifier may be conditional on thesensor data indicating a sensor signal exceeding a limit value; thereby, only relevantevent may be caused to trigger Classification, and energy consumption of the trainedclassifier may be kept at a minimum.

According to embodiments, the trained classifier may be configured as anartificial neural network. The artificial neural network may comprise a plurality oflayers. At least one of the layers may be a convolutional layer. The convolutionallayer may be a one-dimensional convolutional layer configured to convolute atemporally ordered series of motion data.

According to embodiments, the sensor arrangement may comprise at leastone sensor configured to detect a movement of the operator. The at least one sensormay comprise, for example, one or several accelerometers and/or gyroscopes. Alsoe.g. a GNSS receiver, a magnetometer and/or a compass may be suitable fordetecting a movement of the operator. Alternatively or additionally, the sensorarrangement may comprise e.g. a thermometer configured to detect a temperaturechange when stepping out of the operator cabin.

According to embodiments, the trained classifier may be trained to detect thecabin egress event at least partly by detecting that at least a part of the operatorturns about a vertical axis. ln most types of work vehicles, the operator exits thecabin by turning around to face a ladder, handles and footsteps, or similar. Thereby,a typical egress may comprise the operator turning about a vertical axis. Such a turnis also highly suitable for distinguishing an egress motion from motion typical of awork vehicle during operation thereof. For most work vehicle types, the ladder orfootsteps are positioned on a lateral side of the work vehicle, with respect to aforward driving direction of the work vehicle. Thereby, the operator typically turnsabout 40 to 90 degrees about a vertical axis during egress, and the trained classifiermay be trained to detect a turn of this magnitude. The trained classifier may betrained to identify left-side entry, right-side entry, or both.

According to embodiments, the trained classifier may be trained to detect thecabin egress event at least partly by detecting that at least a part of the operatorturns about a horizontal axis. A typical egress may comprise the operator leaninghis/her torso forward, at least initially, during egress, in order to move the bodyweight to a position above the feet.

According to embodiments, the trained classifier may be trained to detect thecabin egress event at least partly by detecting that the operator takes at least onestep down from a floor or seat of the operator cabin to the ground. Such a step isalso highly suitable for distinguishing an egress motion from motion typical of a workvehicle while driving. Typically, any rotation of the operator's body or torso mayprecede stepping down.

According to embodiments, the trained classifier may be trained to detect thecabin egress event at least partly by detecting that the operator takes at least twosteps down from the floor or seat of the operator cabin to the ground. Most typicalwork vehicles comprise at least one intermediate egress footstep. Typically, theintermediate egress footstep is positioned on a lateral side of the work vehicle, withrespect to a forward driving direction of the work vehicle. The work vehicle may alsocomprise e.g. two, three, or four intermediate footsteps, and the trained classifiermay be trained accordingly. The egress detector may be trained to detect severaldifferent numbers of steps, and generate the egress event signal in each situation.

According to embodiments, the safety wearable may be a safety garmentconfigured to be worn on a torso of the operator. Such a garment is well positionedon the body of the operator to provide e.g. visual safety indications. Moreover, ifcombined with movement detection, a high degree of recognizability of the operator'smotion pattern during egress may be obtained, since the torso generally remains in awell-defined position during operation of the work vehicle, and follows a highly typicalroute during egress. The safety garment may be, for example, a vest or a jacket.

According to embodiments, the sensor arrangement may comprise at leastone sensor arranged adjacent to the operator's neck. For example, the at least onesensor may be arranged in a collar of a safety garment. A position adjacent to theoperator's neck provides a well-defined position for all body shapes. Moreover, whencombined with movement detection, a high degree of recognizability of the motionpattern may be obtained.

According to embodiments, the safety wearable may further comprise awarning light configured to enhance visibility of the wearer, i,e, operator, in darkness,wherein the data processing equipment is configured to enable, based on the egressevent signal, the warning light. The safety wearable may further comprise an ambientlight detector, and the data processing equipment may be configured to enable thewarning light based also on a further condition that the ambient light detectorindicates that the ambient light is darker than a threshold level. Thereby,unnecessary switching on of the safety wearable in daytime may be automaticallyavoided.

According to embodiments, the safety wearable may further comprise awireless communication interface configured to wirelessly transmit the egress eventsignal. The egress event signal may be transmitted to the work vehicle, e.g. toautomatically switch on lights of the vehicle or enable a vehicle immobilizer forpreventing theft. Alternatively or additionally, the egress event signal may betransmitted to a remote operator monitoring system, optionally via a wireless relaydevice such as a mobile phone, to thereby warn e.g. the operator's manager that theoperator has exited the vehicle. Exemplary wireless communications interfaces whichare suitable are, e.g., ZigBee and Bluetooth Low Energy.

According to embodiments, the safety wearable may further comprise aGNSS, global navigation satellite system, receiver, wherein the wirelesscommunication interface is configured to, based on the egress event signal,wirelessly transmit the operator's position. Thereby, the operator's manager, or anyoperator support or security functions, may be automatically alerted in case theoperator exits the vehicle in an area marked as unsafe, due to e.g. intenseconstruction vehicle traffic, or in an unexpected area, due to e.g. vehicle hi-jacking orsimilar.

According to embodiments, the safety wearable may further comprise awireless communication interface configured to wirelessly receive a safety wearableactivation scheduling signal, wherein said data processing equipment is configured togenerate said cabin egress event signal based on said safety wearable activationscheduling signal. The safety wearable activation scheduling signal may be receivedfrom e.g. an application running in a mobile phone in wireless communication withthe safety wearable. The data processing equipment may, according to an example,enable activation of egress detection and/or egress signal generation in directresponse to having received the safety wearable activation scheduling signal.Alternatively, the safety wearable activation scheduling signal may comprise daylighttime information indicating e.g. the local time of sunrise and sunset, and the dataprocessing equipment may be configured to set an activation time and/or adeactivation time based on the received daylight time information.

According to embodiments, the trained classifier may further be trained todetect a cabin ingress event, indicating that operator enters or has entered theoperator cabin of the work vehicle, wherein the data processing equipment is furtherconfigured to generate, based on the classification received from the trainedclassifier, a cabin ingress event signal in response to the Classification indicating aningress event. Similar to an egress, an ingress event may be detected by the trainedclassifier being trained to detect the cabin ingress event at least partly by detectingthat at least a part of the operator turns about a vertical and/or horizontal axis, and/orthat the operator takes at least one or more steps up from the ground to a floor or aseat of the operator cabin. Similar to the egress event signal, the ingress event signalmay be used for switching off e.g. a warning light of the wearable, or to wirelesslycommunicate the operator's position or actions to a remote entity.

According to a second aspect, there is provided a computer-implementedmethod of detecting egress from a work vehicle, the method comprising: receivingsensor data from a sensor arrangement; feeding the sensor data as a feature vectorto a trained classifier trained to detect a cabin egress event, the cabin egress eventindicating that an operator exits or has exited an operator cabin of the work vehicle;receiving a classification from the trained classifier; and generating, based on theclassification received from the trained classifier, a cabin egress event signal inresponse to the classification indicating an egress event. The method may beimplemented fully or partly in e.g. a safety wearable as defined hereinabove, in anysuitable portable device, such as a smart phone, or in a combination thereof. Theegress event signal may be transmitted to remote entities, and/or used for e.g.enabling a warning light worn by the operator, as defined hereinabove.

According to a third aspect, there is provided data processing equipmentcomprising at least one processor and memory, configured to carry out the methoddefined hereinabove. The method may be implemented in e.g. a microcontrollerhaving the trained classifier loaded therein. The microcontroller may be positioned ine.g. a collar of a garment.

According to a fourth aspect, there is provided a computer program productcomprising instructions which, when the program is executed on a processor, carriesout the method defined hereinabove.

According to a fifth aspect, there is provided a method of configuring anegress detector of a safety wearable, the method comprising: receiving sensor datafrom a sensor arrangement worn on the body of an operator during egress from awork vehicle; and training a classifier with the sensor data, to obtain a trainedclassifier configured to identify an egress event. The method may be repeated foreach of a plurality of different work vehicles, and/or for each of a plurality of differentOperators, to obtain a training data set representing different operator body shapes,operator behaviours, and/or work vehicle types. The trained classifier may betransferred to the data processing equipment, such as a microcontroller, of a safetywearable.

According to a sixth aspect, there is provided a computer-readable storagemedium having stored thereon or the computer program product or the trainedclassifier defined hereinabove. lt is noted that embodiments of the invention may be embodied by all possiblecombinations of features recited in the c|aims. Further, it will be appreciated that thevarious embodiments described for the devices are all combinable with the methodsas defined in accordance with the different aspects of the present invention, and viceversa.

Brief description of the drawinqs The above, as well as additional objects, features and advantages of thepresent invention, will be better understood through the following illustrative and non-limiting detailed description of preferred embodiments of the present invention, withreference to the appended drawings, where the same reference numerals will beused for similar elements, wherein: Fig. 1 is a side view of a work vehicle according to a first embodiment; Fig. 2 is a side view of the work vehicle of Fig. 1 and an operator of the workvehicle; Fig. 3 is a perspective view of a safety wearable; Fig. 4 is a schematic block diagram illustrating the functional blocks of anelectronics module of the safety wearable of Fig. 3; Fig. 5A is a side view of the work vehicle and operator of Fig. 1 with theoperator seated in a driver's seat of the work vehicle; Fig. 5B is a side view of the work vehicle and operator of Fig. 5A, illustratingthe operator during a first step of egress; Fig. 5C is a side view of the work vehicle and operator of Fig. 5B, illustratingthe operator during a second step of egress; Fig. 5D is a side view of the work vehicle and operator of Fig. 5C, illustratingthe operator during a third step of egress; Fig. 5E is a side view of the work vehicle and operator of Fig. 5D, illustratingthe operator during a fourth step of egress;Fig. 5F is a side view of the work vehicle and operator of Fig. 5E, illustratingthe operator during a fifth step of egress; Fig. 6 is a schematic illustration of functional layers of a trained classifierimplemented in the electronics module of Fig. 4; Fig. 7 is a schematic illustration of a sequence of sensor data fed to thetrained classifier of Fig. 6; Fig. 8 is a side view of a work vehicle according to a second embodiment; Fig. 9 is a flow chart illustrating steps in a method of detecting egress; Fig. 10 is a flow chart illustrating steps in a method of training a classifier foruse in the method of Fig. 9; and Fig. 11 is a perspective view of a data carrier.

All the figures are schematic, not necessarily to scale, and generally only showparts which are necessary in order to elucidate the embodiments, wherein other partsmay be omitted.

Detailed description of the exemplarv embodiments Fig. 1 illustrates the front part of work vehicle embodied as a cargo truck 10,as known in the art, and Fig. 2 illustrates the work vehicle 10 of Fig. 1 along with anoperator, i.e. driver, 12 facing the work vehicle 10. With reference to Fig. 1, the workvehicle 10 has an operator cabin 14, sometimes referred to as a "driver's cabin" orjust "cab", from the interior of which the operator 12 drives the work vehicle. Whendriving the work vehicle 10, the operator/driver 12 is seated in a driver's seat 16, andcontrols the work vehicle 10 by interacting with various controls, such as pedals (notillustrated) and a steering wheel.

The work vehicle 10 has a front end 18, which faces in a forward direction Fwith respect to a typical driving direction of the work vehicle 10, and the operator 12,while seated, faces in the forward direction F. The front end 18 is provided with whiteheadlights 17 and orange turn signal lights 19, which turn signal lights 19 can also beoperated as warning lights. Typically, the operator cabin 14 of a work vehicle 10 hasdoors on its lateral sides to permit the operator 12 to enter and exit the operatorcabin 14; however, for reasons of clarity of illustration, the work vehicle 10 of Fig. 1 isillustrated with its left-side door removed to expose the operator ingress/egressopening 20 normally covered by the door. lt will be appreciated that the operatorcabin 14 may be provided with doors, and the position of a door 22 is lined out by adotted line in Fig. 1. The operator cabin 14 has a floor 24 which may be quite highabove the ground 26; a floor height H of between 100 cm and 200 cm may be typical.ln order to facilitate ingress and egress, a set of intermediate footsteps 28a-c isprovided on a |atera| side of the operator cabin 14. Here, "intermediate" should beconstrued as being vertically positioned between the level of the floor 24 and thelevel of the ground 26. ln the illustrated embodiment, the set of footsteps comprisesthree footsteps, namely, counted from above, a first footstep 28a, a second footstep28b, and a third footstep 28c. The set of footsteps 28a-c may comprise a set of innerfootsteps 28a, which are located inside the door 22, and a set of outer footsteps 28b,28c, which are located outside the door 22. ln the illustrated embodiment, the set ofinner footsteps 28a comprises only the first footstep 28a, whereas the set of outerfootsteps comprises the second and third footsteps 28b, 28c. ln order to further facilitate egress and ingress, the ingress/egress opening 20is provided with an ingress/egress handle arrangement comprising a front handle 30aand a rear handle 30b. The handles 30a, 30b are held by the operator 12 duringingress and egress, and may typically be arranged inside the doorOn his/her torso 11 (Fig. 2), the operator 12 wears a safety wearableconfigured as a safety vest 32, which is illustrated in greater detail in Fig. 3. Thesafety vest 32 comprises a fabric base garment 34 provided with a front zipper 36.On an outer face thereof, the base garment 34 is provided with a set of reflectors,configured as flexible, reflective strips 38, and a set of light sources 40. The reflectivestrips 38 and the light sources 40 are sewn and/or glued to the base garment 34. Thelight sources 40 operate as warning lights enhancing the visibility of thewearer/operator 12 in darkness; by way of example, they may be configured asflexible electroluminescent laminate strips. The safety vest 32 further comprises anelectronics module 42 integrated within the back of the collar 44 of the safety vest 32;its position is schematically indicated in the view of Fig.Fig. 4 schematically illustrates the functional blocks of the electronics module42. The electronics module 42 comprises data processing equipment 46 comprisinga processor 48 in communication with a memory 50. The data processing equipmentmay be implemented using e.g. a generic, commercially available microcontrollerunit, such as an ARM Cortex M4 microcontroller unit. The electronics module 42further comprises a sensor arrangement 52 configured to detect the operator'smovements while wearing the safety vest 32 (Fig. 3). ln the illustrated embodiment,the sensor arrangement 52 is co-located with the data processing equipment 46within the electronics module 42; however, it will be appreciated that the dataprocessing equipment 46 and the sensor arrangement 52 may be arranged inseparate modules at different positions on the safety vest 32. The sensorarrangement 52 may comprise, for example, an accelerometer 54 configured todetect accelerations in the three axes of space. The sensor arrangement may alsocomprise e.g. an ambient light detector 55, a gyroscope 56, a thermometer 57,and/or a GNSS (Global Navigation Satellite System) receiver 58 configured toreceive positioning information from a set of GNSS satellites 60. The data processingequipment 46 and the sensor arrangement 52 are powered by a battery 62, which isalso comprised in the safety vest 32. Furthermore, the sensor arrangement 42 is incommunication with the data processing equipment 46, which enables the sensorarrangement 52 to provide sensor input representative of the operator's 12movements to the data processing equipment 46. The data processing equipment 46processes the sensor input from the sensor arrangement 52, and controls the lightsources 40 in response to specific movement patterns of the operator's torso asidentified by the data processing equipment 46. For the purpose, the data processingequipment 46 has stored thereon a trained classifier 64 trained to detect a cabinegress event, which will be described in the following. The trained classifier 64 maybe configured as an artificial neural network.

As illustrated in Fig. 4, the electronics module may also comprise a wirelessinterface 66 for bidirectional wireless communication with a device external to thesafety vest 32, such as a mobile phone 68 and/or the work vehicleFigs 5A-5F illustrate intermediate steps of an exemplary cabin egress event,i.e. the operator's 12 act of exiting the operator cabin 14 of the work vehicle 10.During the steps illustrated in Figs 5A-F, sensor data representing the cabin egressevent, from the sensor arrangement 52, may be recorded to train the trainedclassifier 64. Alternatively or additionally, during the steps illustrated in Figs 5A-F,sensor data representing the cabin egress event, from the sensor arrangement 52,may be recorded in order to detect or categorize the event at a cabin egress event byprocessing the sensor data through the trained classifier 64. The output receivedfrom the trained classifier may be a classification which identifies the sensor data asrepresenting a cabin egress event with a certain probability or confidence level. lf thedegree of probability exceeds a limit probability, for example a 90% likelihood, thatthe sensor data represents a cabin egress event, the data processing equipmentgenerates a cabin egress event signal, which in turn triggers the activation of the lightsources 40 (Fig. 3) on the safety vestThe exemplary cabin egress event of Figs 5A-5F comprises exemplaryintermediate steps which will now be described in detail with reference to therespective drawings. ln Fig. 5A, the operator 12 is illustrated in his/her normal working positionwhile driving the work vehicleln a first step of egress illustrated in Fig. 5B, the operator 12 leans his/hertorso 11 forward as indicated by the arrow, which moves the operator's body weighttowards a position above the feet. This motion is manifested as a translation of thetorso 11 and the safety vest 32 (Fig. 3) carried thereupon, as well as a rotation abouta horizontal axis perpendicular to the forward direction F. ln a second step of egress illustrated in Fig. 5C, the operator 12 turns about asubstantially vertical axis V to face in a lateral direction perpendicular to the forwarddirection F, grabs the egress handles 30a, 30b, and positions a first foot on the firstfootstep 28a. This motion is manifested as a lateral translation of the torso 11 withrespect to the forward direction F, a rotation about the vertical axis V, and a verticaltranslation downwards when taking the first step. lt will be appreciated that e.g. avertical translation representing a step down will cause the accelerometer 54 (Fig. 4)to indicate an acceleration downwards, followed by an acceleration upwards, whichacceleration upwards is actually a representation of the retardation of the downwardsmotion. Each step downwards manifests itself as such pairs of downwards andupwards accelerations. ln a third step of egress illustrated in Fig. 5D, the operator 12 steps down tothe second footstep 28b by positioning a second foot on the second footstep 28b.This motion is manifested as a vertical translation of the torso 11/safety vest 32downwards when taking the second step, along with a discernibly low level of rotationabout the vertical axis V (Fig. 5C) due to a maintained body posture, facing the dooropening. ln a fourth step of egress illustrated in Fig. 5E, the operator 12 steps down tothe third footstep 28c by moving the first foot to the third footstep 28c. This motion isalso manifested as a vertical translation of the torso 11/safety vest 32 downwardswhen taking the third step, still together with a discernibly low level of rotation aboutthe vertical axis V (Fig. 5C). ln a fifth step of egress illustrated in Fig. 5F, the operator 12 steps down to theground 26 by moving the second foot to the ground 26, and releases the handles30a, 30b. Also this motion is also manifested as a vertical translation of the torso 11/safety vest 32 downwards together with a discernibly low level of rotation aboutthe vertical axis V (Fig. 5C). After having released the egress handles 30a, 30b, theoperator may again turn about a vertical axis to face to continue in any desireddirection.

While receiving the sensor data from the sensor arrangement 52 (Fig. 4), thedata processing equipment 46 continuously feeds the sensor data as a feature vectorto the trained classifier 64 implemented therein, whereby, based on the operatormovements described hereinabove with reference to Figs 5A-5F, or based on asubset of the operator movements of Figs 5A-5F, the trained classifier 64 providesoutput enabling the data processing equipment 46 to determine whether the motionpattern represents an egress event. Once an egress event has been identified, thedata processing equipment 46 enables, i.e. lights up, the light sources 40 (Fig. 3).Optionally, the data processing equipment 46 may be configured to enable the lightsources 40 on the safety vest 32 only based on the further condition that sensor datafrom the ambient light detector 55 indicates that the operator does not exit the cabin14 (Fig. 1) in full daylight. Once an egress event has been identified, the dataprocessing equipment 46 (Fig. 4) also transmits, via the wireless communicationinterface 66, an egress event signal to the work vehicle 10 prompting the workvehicle to the enable the warning lights 19 (Fig. 1), i.e. to set them in a state ofintermittent flashing. This may be done regardless of whether there is daylight or not.Furthermore, the egress event signal may be transmitted to the operator's mobilephone 68 (Fig. 4), or any other suitable wireless communication device, which mayrelay egress information and position information from the GNSS receiver 58 to e.g. acentralized Worker management system remote from the work vehicleAs an alternative to using an ambient light detector 55 (Fig. 4) for determiningwhether to enable the light sources 40, the wireless communication interface 66 mayreceive daylight information, for example the respective local times of sunrise andsunset, from e.g. the mobile phone 68. ln response to having identified an egressevent, the data processing equipment 46 may enable the light sources 40 based onthe further condition that the daylight information indicates that there is presently nodaylight. The processing equipment 46 may occasionally, for example once a day,receive a safety wearable activation scheduling signal indicating the local time ofsunrise and sunset, and may refer to an internal clock for activating and/ordeactivating the egress signal generation based on the sunrise and sunset times.Fig. 6 illustrates an exemplary structure of the trained classifier 64. The trainedclassifier 64 is configured as a deep convolutional neural network, i.e. it comprises aplurality of layers, at least one of which is convolutional. The at least oneconvolutional layer 70 may perform convolution in one dimension, which maytypically be the time dimension. By way of example, the at least one convolutionallayer 70 may be preceded by a batch normalisation layer 72 to optimize the efficiencyof the at least one convolutional layer 70. The at least one convolutional layer 70 maybe succeeded by one or several additional convolutional layers 74. One or severalactivation layers, such as a ReLu (rectified linear unit) layer 75, may be interposedbetween the convolutional layers 70, 74. The convolutional layer(s) 70, 74 may besucceeded by one or several dense, i.e. fully connected, layers 76. Prior to the(respective) fully connected layer(s), pooling layer(s) and/or dropout layer(s) 78 mayreduce the amount of data to be processed by the fully connected layer(s) 76. One orseveral recurrent layers 79, such as LSTM (long short-term memory) layers, mayalso be included to improve handling of time gaps. At the output, an output activationlayer 80, such as a Softmax layer, may be employed for generating the actualclassification. The classification generated at the output of the trained classifier maycomprise an event type, such as an egress event, an ingress event, an "operating thework vehicle" event, or the like. The classification may also be associated with aconfidencelevel Fig. 7 illustrates an exemplary format of a stream of motion sensor data 82 fedto the trained classifier 64. The sensor data received from the sensor arrangement52 may comprise e.g. linear acceleration in three directions x, y, z from theaccelerometer 54, and from the gyroscope 56, angular motion data representingrotation cp, 0, w about the respective axes defined by the three directions x, y, z. Priorto feeding to the trained classifier 64 (Fig. 4), the sensor data may first be pre-processed by noise filtering and averaging to obtain a data rate of e.g. 5 Hz, theoutput of which is illustrated as the stream of time-sliced motion data 82 extending inthe temporal dimension t. The at least one, or first, first convolutional layer 70 (Fig. 6)convolutes over a time window (kernel) 84, which is moved in time in the directionillustrated by an arrow. The time window may, by way of example, be between 1 and20 seconds, for example about five seconds.

The trained classifier 64 of Fig. 6 may further trained to detect a cabin ingressevent, indicating that operator 12 (Fig. 2) enters or has entered the operator cabin 14of the work vehicle 10, and the data processing equipment 46 may be configured togenerate, when the trained classifier 64 indicates an ingress event, a cabin ingressevent signal. The operator 12 may enter the cabin 14 following the steps illustrated inFigs 5A-5F performed in the opposite order, and the trained classifier 64 may betrained accordingly. The cabin ingress event signal may be used for disabling thelight sources 40 of the safety vest 32. The cabin ingress event signal may also bewirelessly communicated to the work vehicle 10 to switch of the warning lights 19(Fig. 1), and to the mobile phone 68 for relaying to a remote entity.

Fig. 8 illustrates a second work vehicle 110 embodied as an excavator, whichis also known in the art. Similar to the truck 10 (Fig. 1), the excavator110 has anoperator cabin 14 provided with a driver's seat (not illustrated), which can be reachedvia a set of intermediate footsteps 128a-b. ln the illustrated embodiment, the set offootsteps 128-b comprises two footsteps, both of which are located outside the door122. The work vehicle 110 is also provided with an ingress/egress handlearrangement comprising a front handle 130 and a rear handle, wherein the rearhandle is hidden behind the door 122 in the view of Fig. 8. Even though an operator'singress/egress trajectory or movement pattern may differ between the work vehicles10, 110 of Figs 1 and 8, it will be appreciated that the trained classifier 64 of Fig 4may be trained to detect an ingress and/or egress event in either of them, or both.

Fig. 9 is a flow chart illustrating the method of detecting egress from any of thework vehicles 10, 110 described hereinabove. The method, which is implemented inthe data processing equipment 46 and thereby constitutes a computer-implementedmethod, comprises the steps: 91: receiving sensor data 82 (Fig. 7) from a sensor arrangement 52 (Fig. 4); 92: feeding the sensor data 82 as a feature vector to a trained classifier 64(Fig. 7); 93: receiving a classification from the trained classifier 64; and 94: generating, based on the classification received from the trained classifier64, a cabin egress event signal in response to the classification indicating an egressevent.

Fig. 10 is a flow chart illustrating a method of configuring the electronicsmodule 42 (Fig. 3) of the safety vest 32 as an egress detector. The methodcomprises the steps: 95: receiving sensor data 82 from a sensor arrangement worn on the body ofan operator 12, such as the sensor data 82 (Fig. 7) from the sensor arrangement(Fig. 4) when the safety vest 32 (Fig. 3) is worn on the body of the operator 12 (Fig.2), during egress from any of the work vehicles 10, 110; and 96: training a classifier with the sensor data 82, to obtain the trained classifierThe method is preferably repeated several times for each of the different workvehicles 10, 110, and for each of a plurality ofdifferent operators 12. The trainedclassifier 64 may be transferred to the data processing equipment 46 of other, similarsafety vests 32, to enable them to detect egress events.

Fig. 11 illustrates a computer-readable storage medium embodied as aCompact Disc 99, having stored thereon the computer program of the dataprocessing equipment 46 (Fig. 4), or the trained classifier 64 (Fig. 6).

The invention has mainly been described above with reference to a fewembodiments. However, as is readily appreciated by a person skilled in the art, otherembodiments than the ones disclosed above are equally possible within the scope ofthe invention as defined by the appended patent claims. ln the claims, the word"comprising" does not exclude other elements or steps, and the indefinite article "a"or "an" does not exclude a plurality. 14

Claims (18)

1. Claims _ A safety wearable (32) configured to be worn by an operator (12) of a work vehicle (10; 110), the safety wearable (32) comprising:a sensor arrangement (52), anddata processing equipment (46) configured to:feed sensor data (82) from the sensor arrangement (52) as afeature vector to a trained c|assifier (64) trained to detect a cabin egressevent, the cabin egress event indicating that the operator (12) exits or hasexited an operator cabin (14) of the work vehicle (10; 110);receive a c|assification from the trained c|assifier (64); andgenerate, based on the c|assification received from thetrained c|assifier (64), a cabin egress event signal in response to thec|assification indicating an egress event_ The safety wearable according to claim 1, wherein the trained c|assifier (64) is implemented in said data processing equipment (46)_ The safety wearable according to any of the preceding claims, wherein the trained c|assifier (64) is configured as an artificial neural network_ The safety wearable according to any of the preceding claims, wherein the sensor arrangement (52) comprises at least one sensor (54, 56, 58)configured to detect a movement of the operator (12)_ The safety wearable according to claim 4, wherein the trained c|assifier (64) is trained to detect the cabin egress event at least partly by detecting that theoperator (12) takes at least one step down from a floor (24) or seat (16) of theoperator cabin (14) to the ground (26)_ The safety wearable according to claim 5, wherein the trained c|assifier (64) is trained to detect the cabin egress event at least partly by detecting that theoperator (12) takes at least two steps down from the floor (24) or seat (16) ofthe operator cabin (14) to the ground (26) 7. The safety wearable according to any of the preceding claims, wherein thesafety wearable (32) is a safety garment configured to be worn on a torso (11)of the operator (12) 8. The safety wearable according to any of the preceding claims, wherein thesensor arrangement (52) comprises at least one sensor (54, 55, 56, 57, 58)arranged adjacent to the operator's (12) neck 9. The safety wearable according to any of the preceding claims, furthercomprising a warning light (40) configured to enhance visibility of the operator(12) in darkness, wherein the data processing equipment (46) is configured toenable, based on the egress event signal, the warning light (40) 10. .The safety wearable according to any of the preceding claims, furthercomprising a wireless communication interface (66) configured to wirelesslytransmit the egress event signal 11. .The safety wearable according to claim 10, further comprising a GNSS, globalnavigation satellite system, receiver (58), wherein the wireless communicationinterface (66) is configured to, based on the egress event signal, wirelesslytransmit the operator's (12) position 12. .The safety wearable according to any of the preceding claims, furthercomprising a wireless communication interface (66) configured to wirelesslyreceive a safety wearable activation scheduling signal, wherein said dataprocessing equipment (46) is configured to generate said cabin egress eventsignal based on said safety wearable activation scheduling signal 13. .The safety wearable according to any of the preceding claims, wherein the trained classifier (64) is further trained to detect a cabin ingress event,indicating that operator (12) enters or has entered the operator cabin (14) ofthe work vehicle (10; 110), wherein the data processing equipment (46) isfurther configured to generate, based on the classification received from thetrained classifier (64), a cabin ingress event signal in response to theclassification indicating an ingress event 14. .A computer-implemented method of detecting egress from a work vehicle (10; 110), comprising: receiving sensor data (82) from a sensor arrangement (52); feeding the sensor data (82) as a feature vector to a trained classifier(64) trained to detect a cabin egress event, the cabin egress event indicatingthat an operator (12) exits or has exited an operator cabin () of the workvehicle (10; 110); receiving a Classification from the trained classifier (64); and generating, based on the c|assification received from the trainedclassifier (64), a cabin egress event signal in response to the c|assificationindicating an egress event 15. Data processing equipment (46) comprising at least one processor (48) andmemory (50), configured to carry out the method of claim16.A computer program product comprising instructions which, when the programis executed on a processor (48), carries out the method according to claim17.A method of configuring an egress detector of a safety wearable (32), comprising: receiving sensor data (82) from a sensor arrangement (52) worn on thebody of an operator (12) during egress from a work vehicle (10; 110); and training a classifier with the sensor data (82), to obtain a trained classifier(64) configured to identify an egress event 18. .A computer-readable storage medium (99) having stored thereon or thecomputer program product of claim 16 or a trained classifier (64) obtained bythe method of claim 17.