Terrain classification method and apparatus

Abstract

Aspects of the present invention relate to a system (1) for processing image data (IMD(n)) to identify a traversable region (TRR) of an off-road terrain (ORT). The traversable region (TRR) represents a region (TRR) of the off-road terrain (ORT) which is traversable by a vehicle (5). The system (1) includes one or more processor (35) collectively configured to receive image data (IMD(n)) representing an image (IMG(n)) of the off-road terrain (ORT). The image data (IMD(n)) is captured by at least one imaging sensor (21) provided on the vehicle (5). The image data (IMD(n)) is processed using an image segmentation model (SGM) to segment the image (IMG(n)) into a plurality of image segments, the plurality of image segments identifying one or more terrain feature (TF(n)) present in the off-road terrain (ORT). In dependence on an operational capability of the vehicle (5), each of the one or more identified terrain feature (TF(n)) is classified as either being traversable (TTR) or non-traversable (UTR). The traversable region (TRR) of the off-road terrain (ORT) is determined by identifying a region of the image (IMG(n)) which excludes any terrain features (TF(n)) classified as being non-traversable (UTR). The system (1) outputs traversable terrain data (TTD) representing the traversable region (TRR) of the off-road terrain (ORT). Aspects of the invention also relate to a vehicle (5); a method (100) of identifying a traversable region (TRR) of an off-road terrain (ORT); a computer-implemented training method for training an image segmentation model; an image segmentation model (SGM) and computer readable instructions.

Description

[0001] TERRAIN CLASSIFICATION METHOD AND APPARATUS

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to a terrain classification method and apparatus. The method and apparatus may be configured to classify off-road terrain. Moreover, the method and apparatus may identify a region of the off-road terrain which is traversable by a vehicle. Aspects of the invention relate to a system for identifying a traversable region of an off-road terrain; a vehicle; a method of identifying a traversable region of an off-road terrain; a computer-implemented training method for training an image segmentation model; an image segmentation model and computer readable instructions.

[0004] BACKGROUND

[0005] It is known to provide a control system in a vehicle to select a subsystem control mode in dependence on a class (or type) of terrain being traversed by the vehicle. The control system may, for example, receive state indicators providing an indication of an operating state of the vehicle. The state indicators are typically received from sensors provided on-board the vehicle to monitor operating parameter(s) of the vehicle. By analysing the state indicators, the control system can identify a subsystem control mode which is suitable for traversing the prevailing terrain. A potential limitation of this approach is that the control system is re-active and can only use vehicle inputs to infer the appropriate subsystem control mode. The control system is unable to characterise terrain features in the path of the vehicle. Rather, the control system can only assess the terrain features as they are encountered by the vehicle.

[0006] It is an aim of the present invention to address one or more of the disadvantages associated with the prior art.

[0007] SUMMARY OF THE INVENTION

[0008] Aspects and embodiments of the invention provide a system for identifying a traversable region of an off-road terrain, a vehicle, a method of identifying a traversable region of an off-road terrain, a computer-implemented training method for training an image segmentation model, an image segmentation model and computer readable instructions as claimed in the appended claims.

[0009] According to an aspect of the present invention there is provided a system for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the system comprising one or more processor collectively configured to: receive image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; process the image data using an image segmentation model to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain; wherein, in dependence on an operational capability of the vehicle, each of the one or more identified terrain feature is classified as either being traversable or non-traversable; determine the traversable region of the off-road terrain by identifying a region of the image which excludes any terrain features classified as being non-traversable; and output traversable terrain data representing the traversable region of the off-road terrain.

[0010] At least in certain embodiments, the image segmentation model is operable to differentiate between terrain features which are traversable or non-traversable. The classification of the or each terrain feature is made in dependence on the operational capability of the vehicle. The operational capability of the vehicle may define an off-road capability of the vehicle. In use, the system processes the image data derived from the at least one imaging sensor and identifies the or each region of the off-road terrain which is traversable by the vehicle. The traversable terrain data output by the system identifies the traversable region of the off-road terrain. The traversable terrain data may be used to generate a graphical representation of the traversable region of the off-road terrain and / or the non- traversable region of the off-road terrain. For example, the traversable region and / or the non-traversable region may be displayed as an overlay provided over the image data. A driver of the vehicle may control the vehicle so as to remain within the traversable region of the off-road terrain while avoiding the non-traversable region. Alternatively, or in addition, one or more vehicle subsystem may be controlled in dependence on the traversable terrain data. The traversable terrain data may be output to a vehicle subsystem controller. For example, the one or more vehicle subsystem may be controlled such that the vehicle traverses the off-road terrain. This may enable autonomous, semi-autonomous or assisted control of the vehicle in an off-road environment.

[0011] The at least one imaging sensor may comprise or consist of an optical camera configured to detect visible light. The optical camera may comprise or consist of a mono-camera or a stereo camera. The or each optical camera may be configured to detect light in a region of the electromagnetic spectrum which is visible to the human eye. The image may be referred to as a visible (optical) image. The image data may be referred to as optical image data.

[0012] The region of the image identified as the traversable region of the off-road terrain excludes any terrain feature classified as being non- traversable. The region of the image identified as the traversable region of the off-road terrain may comprise one or more terrain feature classified as being traversable.

[0013] The classification of the one or more identified terrain feature may comprise determining a surface gradient of the terrain feature. The terrain feature may be classified as being a non-traversable terrain feature in dependence on a determination that the surface gradient is greater than a gradient threshold. The terrain feature may be classified as being a traversable terrain feature in dependence on a determination that the surface gradient is less than the gradient threshold. The surface gradient may be determined relative to surrounding terrain. For example, the image segmentation model may be configured to identify a change in the surface gradient. The gradient threshold may be determined in dependence on an operational capability of the vehicle. For example, the gradient threshold may be defined as being less than or equal to the vehicle approach angle and / or the vehicle departure angle.

[0014] The classification of the one or more identified terrain feature may comprise determining a step-change in height associated with the terrain feature. The terrain feature may be classified as being a non-traversable terrain feature in dependence on a determination that there is an associated step-change in height which is greater than a step-change height threshold. The step-change height threshold may be determined in dependence on an operational capability of the vehicle. For example, the step-change height threshold may be defined in dependence on one or more of the following: wheel diameter; vehicle approach angle; vehicle departure angle; a suspension travel range; and axle articulation. The terrain feature may be identified as comprising a step-change in height relative to surrounding terrain. The step-change in height may comprise an increase in height (+ve), for example a boulder, a crest or a ridge. Alternatively, the step-change in height may comprise a decrease in height (-ve), for example a hole, a trough, a trench or a wheel rut.

[0015] The classification of the one or more identified terrain feature may comprise determining a height of the terrain feature. The terrain feature may be classified as being a non-traversable terrain feature in dependence on a determination that the height is greater than a height threshold.

[0016] The classification of the one or more identified terrain feature may comprise allocating one of a plurality of traversability grades to the or each identified terrain feature. The image segmentation model may determine the traversability grade. The traversability grades may provide an indication of a severity of a terrain feature, for example comprising one or more of the following: a height, a stepchange height and a surface gradient. The or each identified terrain feature may be classified as being traversable if the allocated traversability grade is less than a threshold value; and / or the or each identified terrain feature is classified as being non-traversable if the allocated traversability grade is greater than or equal to a threshold value.

[0017] The image segmentation model may be configured to process the image data to classify each of the one or more identified terrain feature as being either traversable or non-traversable. The image segmentation model may be trained to perform the classification according to the operational capability of the vehicle. More than one image segmentation model may be trained to reflect the operational capabilities of different vehicles. The system may be configured to apply the image segmentation model associated with (or most closely corresponding to) the operational capabilities of the current vehicle.

[0018] Alternatively, or in addition, the classification of the terrain feature may utilise additional sensor data, for example sensor data received from a LIDAR sensor or a radar sensor. The additional sensor data may be used to classify the one or more identified terrain feature.

[0019] The system may be configured to plot a route for the vehicle across part or all of the off-road terrain. The route may be plotted within the determined traversable region of the off-road terrain. The route may avoid or circumnavigate (bypass) the or each terrain feature identified as being non-traversable. The route plotted for the vehicle may comprise traversing one or more terrain feature classified as being traversable. Plotting the route may comprise determining a position of the or each terrain feature classified as being a non- traversable terrain feature in relation to the vehicle. The positioning of the or each terrain feature may be determined by processing the image data. Alternatively, or in addition, positioning may be determined utilising additional sensor data, for example the sensor data received from a LIDAR sensor or a radar sensor.

[0020] The determination of the traversable region of the off-road terrain may comprise determining a separation distance between a first and a second of the identified terrain features classified as being non-traversable. The determination of the traversable terrain comprises determining that the separation distance is greater than a threshold separation. The threshold separation may be defined in relation to a width of the vehicle. For example, the threshold separation may correspondence to the maximum width of the vehicle plus a clearance allowance.

[0021] The classification of the one or more identified terrain feature may comprise identifying that the terrain feature comprises or consists of a wheel rut. The terrain feature may be classified as being traversable in dependence on a determination that the terrain feature comprises or consists of the wheel rut. The traversable terrain may be determined as following the wheel rut.

[0022] The classification of the one or more identified terrain feature as being traversable may comprise defining one or more traversal direction in which the terrain feature is traversable by the vehicle. The terrain feature may be non-traversable in other directions. The traversable terrain data may be generated in dependence on the one or more traversal direction defined for the or each identified terrain feature. The traversable terrain data may be determined to follow the one or more traversal direction defined for the or each identified terrain feature.

[0023] The traversable terrain data may be output to generate a visual representation of the traversable terrain and / or he non-traversable terrain. For example, the traversable terrain data may be output to a display to generate the visual representation. Alternatively, or in addition, the traversable terrain data may be output to control one or more vehicle systems to control dynamic operation of the vehicle, for example to provide autonomous, semi-autonomous or assisted control of the vehicle.

[0024] According to an aspect of the present invention there is provided a system for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the system comprising one or more processor collectively configured to: receive image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; process the image data using an image segmentation model to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain; classify each of the one or more identified terrain feature is classified as either being traversable or non-traversable; determine the traversable region of the off-road terrain by identifying a region of the image which excludes any terrain features classified as being non-traversable; and output traversable terrain data representing the traversable region of the off-road terrain.

[0025] The classification of the one or more identified terrain feature may comprise allocating one of a plurality of traversability grades to the or each identified terrain feature. The image segmentation model may determine the traversability grades. The traversability grades may provide an indication of a severity of a terrain feature, for example comprising one or more of the following: a height, a stepchange height and a surface gradient. The or each identified terrain feature may be classified as being traversable if the allocated traversability grade is less than a threshold value; and / or the or each identified terrain feature is classified as being non-traversable if the allocated traversability grade is greater than or equal to a threshold value. The threshold value may be defined in dependence on an operating capability of the vehicle.

[0026] According to a further aspect of the present invention there is provided a vehicle control system comprising the system described herein. The vehicle control system may comprise at least one imaging sensor for capturing the image data representing an image of the off-road terrain. The vehicle control system may be configured to plot a route for the vehicle to traverse the traversable region of the off-road terrain. The vehicle control system may comprise one or more vehicle subsystem controller for controlling the one or more vehicle subsystems. In use, the one or more vehicle subsystem controller may control the one or more vehicle subsystems such that the vehicle follows the route plotted in dependence on the traversable terrain data. The vehicle control system may thereby implement autonomous, semi-autonomous or assisted control of the vehicle.

[0027] The at least one imaging sensor may comprise or consist of an optical camera configured to detect visible light. The optical camera may comprise or consist of a mono-camera or a stereo camera. The or each optical camera may be configured to detect light in a region of the electromagnetic spectrum which is visible to the human eye. The image may be referred to as a visible (optical) image. The image data may be referred to as optical image data.

[0028] According to a further aspect of the present invention there is provided a vehicle comprising the vehicle control system or the system described herein. According to a further aspect of the present invention there is provided a method for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the method comprising: receiving image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; processing the image data using an image segmentation model to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain; classify each of the one or more identified terrain feature is classified as either being traversable or non-traversable in dependence on an operational capability of the vehicle; determining the traversable region of the off-road terrain by identifying a region of the image which excludes any terrain features classified as being non-traversable; and outputting traversable terrain data representing the traversable region of the off-road terrain.

[0029] The at least one imaging sensor may comprise or consist of an optical camera configured to detect visible light. The optical camera may comprise or consist of a mono-camera or a stereo camera. The or each optical camera may be configured to detect light in a region of the electromagnetic spectrum which is visible to the human eye. The image may be referred to as a visible (optical) image. The image data may be referred to as optical image data.

[0030] According to a further aspect of the present invention there is provided a computer-implemented training method for training an image segmentation model to segment an image to identify terrain features present in an off-road terrain and to classify the identified terrain features in dependence on an operational capability of a vehicle; the method comprising receiving a plurality of training data sets, the training data sets comprising: a first image data representing a plurality of first images, each of the first images representing a first terrain feature, the first image data being labelled as being traversable; a second image data representing a plurality of second images, each of the second images representing a second terrain feature, the second image data being labelled as being non-traversable; and training the image segmentation model to differentiate between terrain features which are traversable and non-traversable.

[0031] The first and second image data may be captured by respective first and second optical cameras configured to detect visible light. The first and second optical cameras may be configured to detect light in a region of the electromagnetic spectrum which is visible to the human eye. Each of the first and second images may be referred to as visible (optical) images. The first image data may be referred to as first optical image data; and the second image data may be referred to as second optical image data.

[0032] According to a further aspect of the present invention there is provided a computer-implemented training method for training an image segmentation model to segment an image to identify terrain features present in an off-road terrain and to classify the identified terrain features; the method comprising receiving a plurality of training data sets, the training data sets comprising: a first image data representing a plurality of first images, each of the first images representing a first terrain feature, the first image data being labelled as having a first traversability grade; a second image data representing a plurality of second images, each of the second images representing a second terrain feature, the second image data being labelled as having a second traversability grade; and training the image segmentation model to differentiate between terrain features having the first and second traversability grades.

[0033] The first and second traversability grades are different from each other. The method may be performed to train the image segmentation model to differentiate between terrain features having a plurality of different traversability grades. For example, the training data sets may comprise a third image data representing a plurality of third images, each of the third images representing a third terrain feature, the third image data being labelled as having a third traversability grade. Additional image data may be provided representing terrain features having further traversability grades. The image data may be labelled manually, for example by an operator. The resulting image segmentation model may be used to identify a terrain feature and to allocate a traversability grade to the identified terrain feature.

[0034] According to a further aspect of the present invention there is provided an image segmentation model trained using the method(s) described herein. According to a further aspect of the present invention there is provided a system for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the system comprising one or more controller configured to implement an image segmentation model as described herein to process image data generated by an imaging sensor, to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain.

[0035] According to a further aspect of the present invention there is provided computer readable instructions which, when executed by a computer, are arranged to perform the method described herein.

[0036] According to a further aspect of the present invention there is provided a method for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the method comprising: receiving image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; using an image segmentation model generated using the computer-implemented training method described herein to process the image data to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain. The image segmentation model may be trained using the method(s) described herein.

[0037] The method may comprise classifying each of the one or more identified terrain feature as either being traversable or non-traversable in dependence on an operational capability of the vehicle. The traversable region of the off-road terrain may be determined by identifying a region of the image which excludes any terrain features classified as being non-traversable. The method may comprise outputting traversable terrain data representing the traversable region of the off-road terrain.

[0038] Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, in the claims and / or in the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and / or features of any embodiment can be combined in any way and / or combination, unless such features are incompatible. The applicant reserves the right to change any originally filed claim or file any new claim accordingly, including the right to amend any originally filed claim to depend from and / or incorporate any feature of any other claim although not originally claimed in that manner.

[0039] BRIEF DESCRIPTION OF THE DRAWINGS

[0040] One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0041] Figure 1 shows a schematic representation of a vehicle incorporating a system for identifying a traversable region of an off-road terrain in accordance with an embodiment of the present invention;

[0042] Figure 2 shows a schematic representation of an image processing system for implementing a segmentation model to identify terrain features;

[0043] Figure 3 is a block diagram representing a method of identifying the traversable region in accordance with an embodiment of the present invention;

[0044] Figures 4A and 4B show a first example of the operation of the segmentation model to identify a traversable region within the off-road terrain;

[0045] Figures 5A and 5B show a second example of the operation of the segmentation model to identify a traversable region within the offroad terrain;

[0046] Figures 6A and 6B show a third example of the operation of the segmentation model to identify a traversable region within the off-road terrain;

[0047] Figures 7A and 7B show a fourth example of the operation of the segmentation model to identify a traversable region within the offroad terrain; and

[0048] Figure 8 illustrates a computer-implemented method of training a segmentation model for use in the system according to an embodiment of the present invention.

[0049] DETAILED DESCRIPTION

[0050] A system 1 and method 100 for identifying a traversable region TRR of an off-road terrain (denoted generally by the reference numeral ORT) in accordance with an embodiment of the present invention is described herein with reference to the accompanying Figures.

[0051] The system 1 is provided in a vehicle 5 comprising four (4) wheels (not shown). The vehicle 5 in the present embodiment is an offroad vehicle, such as a sports utility vehicle (SUV), a utility vehicle or a truck. The system 1 may be installed in other types of vehicle 5. The system 1 is configured to analyse the off-road terrain ORT to identify terrain features TF(n)-n and to classify the terrain features TF(n)-n as being traversable or non-traversable. As described herein, the system 1 is configured to identify the traversable region TRR in dependence on an operational capability of the vehicle 5.

[0052] The vehicle 5 comprises at least one electric drive unit 7 configured to drive two (2) or four (4) of the wheels. The electric drive unit 7 is supplied with electrical energy stored in a traction battery 9. One or more inverter 11 is provided for converting the direct current (DC) from the traction battery 9 into alternating current (AC) which is supplied to the at least one electric drive unit 7. The vehicle 5 is a battery electric vehicle (BEV) in the present embodiment. It will be understood that the system 1 and the method 100 described herein are applicable to other types of vehicle 5, such as a hybrid electric vehicle (H EV), a plug-in hybrid electric vehicle (PHEV) or an internal combustion engine (ICE) vehicle. The system 1 is configured to monitor a region of terrain proximal to the vehicle 5, typically in front of the vehicle 5. The vehicle 5 comprises one or more imaging sensor 21 configured to capture image data IMD(n) representing an image IMG(1 ). The vehicle 5 in the present embodiment comprises one or more first imaging sensor 21 configured to capture first image data IMD(1) representing a first image I MG(1 ). The first imaging sensor 21 has a direct line-of-sight to the terrain ORT, preferably reducing or avoiding reflections (for example in a side mirror or rear-view mirror). The system 1 described herein may have a dedicated first imaging sensor 21 to capture the first image data I M D(1 ). It will be understood that the first imaging sensor 21 may also be shared with other vehicle systems. The first imaging sensor 21 in the present embodiment is mounted in an elevated position, for example at the top of a front windshield of the vehicle 5. The first imaging sensor 21 may be mounted in other locations on the vehicle 5. The first imaging sensor 21 may have different orientations / directions to the arrangements illustrated herein. The first image IMG(1) is a dynamic image which changes with respect to time. The vehicle 5 is described herein as comprising one said first imaging sensor 21, although it will be appreciated that this is merely illustrative. The first imaging sensor 21 in the present embodiment is an optical camera configured to detect visible light. The first imaging sensor 21 is configured to detect light in the portion of the electromagnetic spectrum that is visible to the human eye. The first image IMG(1) is a visible (optical) image I MG(1 ).

[0053] As illustrated in Figure 1 , the first imaging sensor 21 has a first field of view FOV1. The first field of view FOV1 extends in front of the vehicle 5 such that the first image I MG(1 ) represents a scene to a front of the vehicle 5. The first imaging sensor 21 is a mono-camera in the present embodiment. In a variant, the imaging sensor 21 may comprise a stereo camera for capturing stereo images, for example to determine a distance (range) to features. The vehicle 5 may comprise one or more second imaging sensor 23 configured to capture second image data IMD(2) representing a second image IMG(2). The second image IMG(2) may be comprise or consist of a three- dimensional representation of the off-road terrain ORT, for example in the form of a point cloud. The one or more second imaging sensor 23 may be a different type of sensor than the first imaging sensor 23. The second imaging sensor 23 may comprise one or more of the following: a LIDAR sensor and a radar sensor. The first image I MG(1 ) and the second I MG(2) may represent the same (or a partially over-lapping) region of the off-road terrain ORT. Both the first image data IMD(1) and the second image data I MD(2) may be analysed to identify the traversable region of the off-road terrain ORT. The first and second image data IMD(1), IMD(2) may be processed to identify the same terrain feature(s) TF(n) in each of the first and second images I MG(1 ), IMG(2).

[0054] The system 1 comprises an image processing system 31 for processing the first image data IMD(1) received from the first imaging sensor 21. The image processing system 31 in the present embodiment implements a segmentation model SGM. The segmentation model SGM is configured to segment the first image IMG(1) into a plurality of first semantic areas IMS-n. A computer-implemented training method and system for training the segmentation model SGM in accordance with an embodiment of the present invention is described herein.

[0055] As shown in Figure 2, the image processing system 31 comprises one controller 33, although it will be appreciated that this is merely illustrative. The controller 33 comprises processing means 35 and memory means 37. The processing means 35 may be one or more electronic processing device 35 which operably executes computer-readable instructions. The memory means 37 may be one or more memory device 37. The memory means 37 is electrically coupled to the processing means 35. The memory means 37 is configured to store instructions, and the processing means 35 is configured to access the memory means 37 and execute the instructions stored thereon. When executed, the instructions cause the controller 33 to perform the method(s) described herein. The controller 33 comprises an input means 39 and an output means 41. The input means 39 comprises an electrical input 39 of the controller 33. The input means 39 is configured to receive the first image data IMD(1) representing the first image IMG(1). The input means 39 may optionally be configured to receive the second image data IMD(2) representing the second image IMG(2). The output means 41 may comprise an electrical output 41. The output 41 is arranged to output a traversable terrain signal SG1. The traversable terrain signal SG1 is an electrical signal providing an indication of the region TRR of the off-road terrain ORT identified as being traversable by the vehicle 5. The traversable terrain signal SG1 may providing an indication of one or more region UTR of the off-road terrain ORT identified as being non-traversable by the vehicle 5. The traversable terrain signal SG1 comprises traversable terrain data TTD defining the traversable region TRR of the off-road terrain ORT. The traversable terrain signal SG1 is output to a display screen 43 to provide a graphical representation of the traversable terrain TRR. For example, the traversable terrain TRR may be displayed as an overlay on the first image IMG(1) captured by the first imaging sensor 23.

[0056] As described herein, the image processing system 31 is configured to process the first image data IMD(1) to segment the first image IMG(1) into one or more first image segments IMS-n. The segmentation model SGM in the present embodiment segments the first image IMG(1) to determine a plurality of the first image segments IMS-1. The segmentation model SGM identifies parts of the first image IMG(1) having the same semantic classification. The or each first image segment IMS-n represents a semantic area within the first image I MG(1 ). The first image segments IMS-n correspond to respective terrain features TF(n) present in the off-road terrain ORT represented by the first image IMG(1). The segmentation model SGM is configured to segment the first image IMG(1) on a per-pixel basis such that the segmentation is performed at a pixel level. The process comprises classifying each pixel in the first image IMG(1) as having one of a plurality of semantic classifications. The pixels in the first IMG(1) classified as having the same semantic classification are identified as belonging to the same first image semantic area IMS(n). Alternatively, the segmentation model SGM may segment the first image I MG( 1 ) based on groups or clusters of pixels. This may reduce the computational overhead in performing segmentation of the first image IMG(1).

[0057] The segmentation model SGM segments the first image IMG(1) into a plurality of classes. The classes each represent semantically different classes of pixels. In the present embodiment, the segmentation model SGM segments the first image IMG(1) into two (2) classes. The segmentation provides a probability of each pixel belonging to each of the predefined classes. By way of example, the segmentation may indicate that a particular pixel has a probability of 25% of belonging to a first class and a probability of 55% of belonging to a second class. The semantic classes are referred to therein as terrain classes TCL. The segmentation model SGM determines a terrain class TCL for each of the first image segments IMS-n. Each first image segment IMS-n is classified as being either traversable or non-traversable. The terrain classes TCL identify each of the first image segment IMS-n as corresponding to either a traversable terrain feature TF(n) or a non-traversable terrain feature TF(n). The (or each) traversable terrain feature TF(n) can be traversed by the vehicle 5; and the (or each) non-traversable terrain feature TF(n) cannot be traversed by the vehicle 5. The (or each) non-traversable terrain feature TF(n) thereby represents an obstacle or obstruction which cannot be traversed by the vehicle 5. At least in certain embodiments the classification of the or each terrain feature TF(n) as being traversable or non-traversable is performed in dependence on the operational capability of the vehicle 5.

[0058] In a variant, the segmentation model SGM may segment the first image IMG(1) into more than two classes. In a further variant, the segmentation model SGM may segment the first image I MG(1 ) into a single class. In this arrangement, the segmentation model SGM may classify each pixel in the first image IMG(1) as either being in the class or being unclassified. The class represents one of traversable and non-traversable. The inverse of the class is identified as belonging to the other one of the traversable and non- traversable. The segmentation model SGM is configured to process the first image data I MD(1 ) to determine characteristics of the or each terrain feature TF(n). The segmentation model SGM may, for example, determine one or more of the following characteristics of the or each terrain feature TF(n): i. A surface gradient (or slope angle) of the terrain feature indicating a gradient of one or more surface of the terrain feature TF(n). ii. A change in gradient approaching or departing from the terrain feature TF(n). iii. A step-change (either positive (+ve) or negative (-ve)) in height associated with the terrain feature TF(n), for example a height of a rock step. iv. A height of the terrain feature TF(n) (relative to the vehicle 5 or a reference plane). v. A width of the terrain feature TF(n), for example a width of a wheel rut or a channel. vi. A profile (shape) of the terrain feature TF(n). vii. An orientation of the terrain feature TF(n), for example an orientation of a wheel rut or a channel. viii. A composition of the terrain feature TF(n), for example to differentiate between different terrain classes.

[0059] In the present embodiment, the first image data IMD(1) is captured by the first imaging sensor 21 . The first imaging sensor 21 is an optical camera for capturing visible light. The segmentation model SGM can process the first image data I MD(1 ) to determine the one or more characteristic of the or each terrain feature TF(n). The first imaging sensor 21 is a mono camera in the present embodiment. In a variant, the imaging sensor 21 may comprise a stereo camera for capturing stereo images. The processing of the first image data IMD(1) captured by the stereo camera would facilitate determination of the one or more characteristic of the or each terrain feature TF(n) with improved accuracy. For example, the first image data I MD(1 ) captured by the stereo camera may facilitate determination of a depth and / or a perspective of the or each terrain feature TF(n). At least in certain embodiments, the one or more characteristic may be determined in dependence on data derived from the one or more second imaging sensor 23 configured to capture second image data IMD(2). The one or more second imaging sensor 23 may comprise a LIDAR sensor which uses ultraviolet, visible, or near-infrared light to image objects. Alternatively, or in addition, the one or more second imaging sensor 23 may comprise a radar sensor. The one or more characteristic may be determined in dependence on the first image data IMD(1) captured by the first imaging sensor 21 in combination with the second image data IMD(1) captured by the one or more second imaging sensor 23. For example, the segmentation model SGM may process the image data IMD(1) to identify the one or each terrain feature TF(n). The segmentation model SGM may identify one or more terrain feature TF(n). The second image data IMD(2) from the second imaging sensor 23 may be used to determine the one or more characteristic. It will be understood that the one or more characteristic may be determined exclusively through analysis of the first image data I MD(1 ).

[0060] If two or more terrain features TF(n) are identified, the segmentation model SGM may also consider the relationship between the terrain features TF(n). For example, the segmentation model SGM may determine a separation distance between adjacent terrain features.

[0061] The operational capability of the vehicle 5 may be determined in dependence on one or more vehicle parameter VPS. The one or more vehicle parameter VPS may, for example, comprise one or more of the following: i. A driveline configuration (for example front-wheel drive, rear-wheel drive or four-wheel drive). ii. An approach angle of the vehicle 5. iii. A departure angle of the vehicle 5. iv. A breakover (ramp over) angle of the vehicle 5. v. A ride height of the vehicle 5. vi. A suspension type of the vehicle 5 (air spring or steel spring). vii. A suspension travel range of the vehicle 5. viii. An axle articulation range of the vehicle 5. ix. A size (diameter and / or width) of the wheels on the vehicle 5. x. A tyre rating of the vehicle 5 (for example, one or more of on-road tyre rating and off-road tyre rating). xi. A presence / absence of A transfer case (to select one of A high drive mode and A low drive mode). xii. An external dimension (such as A width and / or A height) of the vehicle 5. xiii. A wheel track of the vehicle 5 (i.e. , A transverse distance between the wheels). xiv. A wheel base of the vehicle 5 (i.e., A longitudinal distance between the wheels).

[0062] The operational capability of the vehicle 5 may be modelled in dependence on the one or more vehicle parameter VPS, for example based on a computational simulation and / or empirical (test) data. The one or more vehicle parameter VPS may be specific to a particular type or model of the vehicle 5. Alternatively, the one or more vehicle parameter VPS may vary depending on the specification of a particular vehicle 5. The one or more vehicle parameter VPS are used to determine a threshold value for assessing whether the terrain feature TF(n) is traversable or non-traversable.

[0063] The approach angle and / or the departure angle of the vehicle 5 may be used to determine a gradient threshold. The gradient threshold defines a maximum change in gradient (either approaching or departing from) the terrain feature TF(n). If the segmentation model SGM determines that the change in gradient associated with a terrain feature TF(n) is greater than the gradient threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as non-traversable. If the segmentation model SGM determines that the change in gradient associated with a terrain feature TF(n) is less than the gradient threshold, the segmentation model SGM classifies the terrain class T CL for that terrain feature TF(n) as traversable. A similar approach may be taken to determine a breakover threshold in dependence on the breakover angle of the vehicle 5. If the segmentation model SGM determines that a breakover angle associated with a terrain feature TF(n) is greater than the breakover threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as non-traversable.

[0064] The wheel base and / or the wheel track of the vehicle 5 may be used to determine a gradient threshold. The gradient threshold defines an upper threshold for the surface gradient of the terrain feature TF(n) that can be traversed by the vehicle 5. If the segmentation model SGM determines that the surface gradient associated with a terrain feature TF(n) is greater than the gradient threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as non-traversable. If the segmentation model SGM determines that the surface gradient associated with a terrain feature TF(n) is less than the gradient change threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as traversable. The gradient threshold may be direction dependent, for example dependent on a direction of travel of the vehicle 5 over the terrain feature TF(n) relative to a direction of the surface gradient. Different gradient thresholds may be defined depending on whether the vehicle 5 is travelling up or down the slope.

[0065] The approach angle and / or the departure angle of the vehicle 5 may be used to determine a gradient change threshold. The gradient change threshold defines an upper threshold for the change in gradient that can be traversed by the vehicle 5, either approaching or departing from a terrain feature TF(n). If the segmentation model SGM determines that the change in gradient associated with a terrain feature TF(n) is greater than the gradient change threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as non-traversable. If the segmentation model SGM determines that the change in gradient associated with a terrain feature TF(n) is less than the gradient change threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as traversable. A similar approach may be taken to determine a breakover threshold in dependence on the breakover angle of the vehicle 5.

[0066] A step-change height threshold (positive and / or negative) may be determined in dependence on one or more of the vehicle parameter VPS, such as one or more of the suspension type, suspension travel range, axle articulation, ride height and wheel size. The stepchange height threshold defines an upper threshold for a stepped change in height that can be traversed by the vehicle 5. The stepped change in height is a discrete or localised change in height, for example formed by a rock step or the like. The change in height may be positive (i.e., higher than the surrounding terrain) or negative (i.e., lower than the surrounding terrain). If the segmentation model

[0067] SGM determines that the change in height of the terrain feature TF(n) is less than the step-change height threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as traversable. If the segmentation model SGM determines that the change in height of the terrain feature TF(n) is greater than the step-change height threshold, the segmentation model SGM classifies the terrain class TCL for that terrain feature TF(n) as non-traversable. The step-change height threshold may be direction dependent, for example dependent on a direction of travel of the vehicle 5 over the terrain feature TF(n). Different step-change height thresholds may be defined depending on whether the vehicle 5 is travelling up or down the step.

[0068] A separation threshold may be determined in dependence on the width of the vehicle 5. If the separation distance between adjacent terrain features TF(n) is less than the threshold separation, the terrain class TCL for the region between the adjacent terrain features TF(n) is determined by the segmentation model SGM as being non-traversable. A wheel track threshold may be determined in dependence on the wheel track of the wheels of the vehicle 5. If the terrain features TF(n) are in the form of channels or wheel ruts, the segmentation model SGM may compare the separation distance to the wheel track threshold. If the separation distance corresponds to the wheel track threshold, the terrain class T CL for the region between the adjacent terrain features TF(n) is determined by the segmentation model SGM as being traversable. If the separation distance is less than or greater than the wheel track threshold, the terrain class TCL for the region between the adjacent terrain features TF(n) is determined by the segmentation model SGM as being non-traversable.

[0069] It will be understood that the classification of the or each terrain feature TF(n) is dependent on the one or more vehicle parameter VPS. The one or more vehicle parameter VPS may be applied as a weight (or bias) to adjust the classification of the or each terrain feature TF(n). The determination of the or each terrain class TCL may, for example, be modified in dependence on the one or more vehicle parameter VPS. Alternatively, the segmentation model SGM may be selected in dependence on the one or more vehicle parameter VPS. For example, one of a plurality of segmentation models SGM may be selected in dependence on the one or more vehicle parameter VPS. The plurality of segmentation models SGM may each correspond to a particular type or configuration of the vehicle 5, as characterised by the one or more vehicle parameter VPS. The selected segmentation model SGM may be installed on-board the vehicle 5 to be executed locally by the image processing system 31.

[0070] In a variant, the segmentation model SGM may define a plurality of terrain classes TCL to classify the terrain features TF(n), for example to define three (3), four (4), five (5) or more terrain classes TCL. The terrain classes TCL may each be defined by a value in a numerical scale, for example ranging from zero (0) to ten (10) inclusive or as a percentage (%) value. The ability of the vehicle 5 to traverse the terrain feature TF(n) may be directly or inversely proportional to the numerical value defined by the terrain class TCL. For example, a terrain class TCL having a low numerical value may indicate that traversal of the terrain feature TF(n) can be performed with relative ease; whereas a higher numerical value may indicate that traversal of the terrain feature TF(n) is more challenging or that the terrain feature TF(n) is non-traversable. A terrain class threshold may be defined to determine whether each of the terrain classes TCL is traversable or non-traversable by the vehicle 5. A terrain class TCL which is less than or equal to the terrain class threshold may be defined as being traversable for the vehicle 5; and a terrain class TCL which is greater than the terrain class threshold may be defined as being non-traversable for the vehicle 5. The terrain class threshold may be defined in dependence on the operational capability of the vehicle 5. In the example of a numerical scale ranging from zero (0) to ten (10) inclusive, a first vehicle having a first set of vehicle parameters VPS may have a first terrain class threshold of four (4); and a second vehicle having a second set of vehicle parameters VPS may have a second terrain class threshold of six (6). A terrain feature TF(n) having a terrain class TCL of five (5) would be classified as non-traversable for the first vehicle but would be classified as traversable for the second vehicle. The terrain class threshold may be determined in respect of the one or more vehicle parameter VPS.

[0071] The input means 39 may optionally also receive a vehicle attitude data ATD1 indicating an attitude of the vehicle 5. The vehicle attitude data ATD1 may, for example, indicate a pitch angle and / or a roll angle of the vehicle 5 relative to a horizontal plane. The vehicle attitude data ATD1 can be used in the processing of the first image data IMD(1) and / or the second image data IMD(2). The vehicle attitude data ATD1 may facilitate determination of the pitch angle and / or the roll angle of the first imaging sensor 21 relative to the horizontal plane, thereby facilitating analysis of terrain features TF(n) present in the first image I MG( 1 ). The vehicle attitude data ATD1 may be received from an inertial measurement unit (I MU) 45 provided onboard the vehicle 5. The IMU 45 may comprise one or more accelerometer and / or one or more gyroscope for determining the attitude of the vehicle 5. The processing of the first image data I MD( 1 ) may be performed in dependence on the vehicle attitude data ATD1. The vehicle attitude data ATD1 may facilitate classification of the (or each) terrain feature TF(n) present in the first image IMG(1). For example, the vehicle attitude data ATD1 may facilitate determination of a surface gradient (slope angle) of the terrain feature(s) TF(n).

[0072] The input means 39 may optionally also receive a vehicle position data POD1 indicating a geospatial position and / or orientation of the vehicle 5. The geospatial position and / or orientation of the vehicle 5 may facilitate determination of topographic data representing terrain contours proximal to the vehicle 5. The topographic data may, for example, be stored in the memory means 37 or a storage device (not shown) provided onboard or offboard the vehicle 5, for example in a remote (cloud) server. The processing of the first image data IMD(1) may be performed in dependence on the vehicle position data POD1 and / or the topographic data. The topographic data may facilitate classification of the (or each) terrain feature TF(n) present in the first image IMG(1). For example, the topographic data may facilitate determination of a surface gradient (slope angle) or a change in height of the terrain feature(s) TF(n).

[0073] As described herein, the traversable terrain signal SG1 comprises the traversable terrain data TTD defining the traversable region TRR of the off-road terrain ORT. It may be desirable to plot a route RT 1 (or path) for the vehicle 5 to traverse the off-road terrain ORT. The system 1 may be configured to plot the route RT 1 within the traversable region TRR identified within the off-road terrain ORT. The system 1 may, example, plot the route RT1 from a current location of the vehicle 5 to a waypoint or destination identified in the offroad terrain ORT. The waypoint or destination may be defined by a user of the vehicle 5 or may be identified by another vehicle subsystem. Alternatively, or in addition, the system 1 may plot the route RT 1 from a first waypoint to a second waypoint or destination identified in the off-road terrain ORT. The system 1 is configured to generate the route RT1 such that movement of the vehicle 5 is constrained to the traversable region TRR identified within the off-road terrain ORT. The system 1 generates a route RT1 to ensure that the vehicle 5 does not traverse or impinge on any regions of the off-road terrain 5 identified as being non-traversable. A set of route data RTD representing the route RT 1 may be output from the system 1 to the display screen 43. The route RT 1 may be displayed on the display screen 43 to help a user control the vehicle 5 to reach the waypoint or destination. Alternatively, the route data RTD may be used by other vehicle subsystems to control the vehicle 5 to follow the route RT 1 , for example to provide autonomous, semi- autonomous or assisted control of the vehicle 5. For example, the route data RTD may be used to control a vehicle steering subsystem to adjust a steering angle of the vehicle 5. The vehicle steering subsystem may be controlled to steer the vehicle 5 along the route RT 1. The route data RTD may be output to a control system for the electric drive unit 7 (or an engine control unit) to control a driving torque to propel the vehicle 5.

[0074] Figure 3 illustrates a method 100 according to an embodiment of the invention. The method 100 is a method of identifying a traversable region TRR of an off-road terrain ORT. The method 100 may be performed by the system 1 described herein. In particular, the memory 37 may comprise computer-readable instructions which, when executed by the processor 35, perform the method 100 according to an embodiment of the invention.

[0075] The method 100 will be described with reference to the vehicle 5 situated in the off-road terrain ORT. The method 100 is initiated (BLOCK 105). The method 100 comprises receiving first image data IMD(1) representing a first image IMG(1 ) of the off-road terrain (BLOCK 110). The first image data IMD(1) is captured by the first imaging sensor 21 provided on the vehicle 5 in the present embodiment. The first image IMG(1) comprises a scene in front of the vehicle 5. The first image IMG(1) is processed using the segmentation model SGM (BLOCK 115). The segmentation model SGM segments the first image I MG(1 ) into one or more first image segments IMS-n. The one or more first image segments IMS-n correspond to terrain features TF(n) occurring in the off-road terrain ORT. In dependence on an operational capability of the vehicle, each of the one or more identified terrain feature TF(n) is classified by the segmentation model SGM as either being traversable or non-traversable (BLOCK 120). The traversable region TRR of the offroad terrain ORT is identified within the first image IMG(1) (BLOCK 125). The traversable region TRR corresponds the or each region of the first image IMG(1) which excludes any terrain features TF(n) classified as being non-traversable. The traversable region TRR may comprise one or more terrain feature TF(n) classified as being traversable. A set of traversable terrain data is output (BLOCK 130). The traversable terrain data represents the traversable region TRR of the off-road terrain ORT. The method may optionally comprise plotting a route RT 1 within the bounds of the traversable region TRR (BLOCK 135). For example, a waypoint or a destination may be set for the vehicle 5, either by a user or a vehicle control system. The route RT 1 may be plotted from the current location of the vehicle 5 to the waypoint or the destination. The method may comprise controlling the vehicle 5 to traverse the off-road terrain ORT by following the route RT 1 plotted within the traversable region TRR (BLOCK 140). The method 100 ends (BLOCK 145).

[0076] A first example of the operation of the segmentation model SGM to identify a traversable region TRR within a section of off-road terrain ORT will now be described with references to Figures 4A and 4B. A first image I MG(1 ) comprising the off-road terrain ORT is captured by the first imaging sensor 21. The first image IMG(1) is shown in Figure 4A by way of example. The off-road terrain ORT in this example comprises an un-metalled surface extending through a desert landscape comprising sand banks and dunes. The un-metalled surface is relatively flat and forms an off-road trail for vehicles. The un-metalled surface is composed of vehicle tracks. There are no formal markings, indicia or street furniture to mark the boundary of the un-metalled surface. A lead vehicle ahead of the host vehicle 5 is visible in the first image I MG(1 ). The terrain features TR identified by the segmentation model SGM include sand banks TF(1), TF(2) formed on opposing sides of the un-metalled surface. The sandbanks have surfaces sloping downwardly towards the un-metalled surface. On the right hand side, the sand bank TF(2) extends partway across the un-metalled surface. The first image data IMD(1) representing the first image IMG(1) is supplied to the segmentation model SGM for processing. The segmentation model SGM segments the first image I MG(1 ) to identify the terrain features TF(n). The terrain features TF(n) are classified as being traversable or non-traversable. The classification is performed in dependence on the operational capabilities of the vehicle 5. The classification may, for example, reflect the off-road capability of the vehicle 5. A first segmented image I MG(1 )(S) corresponding to the first image I MG(1 ) is shown in Figure 4B. The segmentation model SGM identifies the traversable region TRR of the off-road terrain ORT. Alternatively, or in addition, the segmentation model SGM identifies the non-traversable region(s) UTR of the off-road terrain ORT. As described herein, a route RT1 may optionally be plotted within the bounds of the traversable region TRR identified by the segmentation model SGM.

[0077] A second example of the operation of the segmentation model SGM to identify a traversable region TRR within a section of off-road terrain ORT will now be described with references to Figures 5A and 5B. A first image IMG(1) comprising the off-road terrain ORT is captured by the first imaging sensor 21. The first image I MG(1 ) is shown in Figure 5A by way of example. The off-road terrain ORT in this example comprises an un-metalled surface extending through a desert landscape comprising sand banks and dunes. The unmetalled surface is relatively flat and forms an off-road trail for vehicles. The un-metalled surface is composed of vehicle tracks. There are no formal markings, indicia or street furniture to mark the boundary of the un-metalled surface. A lead vehicle ahead of the host vehicle 5 is visible in the first image IMG(1). The terrain features TR identified by the segmentation model SGM comprise an inclined surface TF(1) on the left hand side of the un-metalled surface sloping downwardly away from the un-metalled surface and which hosts some vegetation; and a series of sand banks TF(2) formed on the right-hand side of the un-metalled surface. The sandbanks TF(2) slope downwardly towards, and onto the un-metalled surface. The first image data IMD(1) representing the first image IMG(1) is supplied to the segmentation model SGM for processing. The segmentation model SGM segments the first image IMG(1) to identify the terrain features TF(n). The terrain features TF(n) are classified as being traversable or non-traversable. The classification is performed in dependence on the operational capabilities of the vehicle 5. The classification may, for example, reflect the off-road capability of the vehicle 5. A first segmented image IMG(1)(S) corresponding to the first image IMG(1) is shown in Figure 5B. The segmentation model SGM identifies the traversable region TRR of the off-road terrain ORT. Alternatively, or in addition, the segmentation model SGM identifies the non-traversable region(s) UTR of the off-road terrain ORT. As described herein, a route RT 1 may optionally be plotted within the bounds of the traversable region TRR identified by the segmentation model SGM.

[0078] A third example of the operation of the segmentation model SGM to identify a traversable region TRR within a section of off-road terrain ORT will now be described with references to Figures 6A and 6B. A first image IMG(1) comprising the off-road terrain ORT is captured by the first imaging sensor 21. The first image I MG(1 ) is shown in Figure 6A by way of example. The off-road terrain ORT in this example comprises an un-metalled surface extending between rows of vegetation (trees or bushes). The vegetation on the lefthand side is labelled as the terrain feature TF(1); and the vegetation on the right-hand side is labelled as the terrain feature TF(2). The un-metalled surface is formed of sand. The un-metalled surface is relatively flat and forms an off-road trail for vehicles. There are no formal markings, indicia or street furniture to mark the boundary of the un-metalled surface. The terrain features TR identified by the segmentation model SGM comprise the rows of vegetation on the left- and right-hand sides of the un-metalled surface. The vegetation extends partway across the un-metalled surface in places. The first image data I MD(1 ) representing the first image I MG(1 ) is supplied to the segmentation model SGM for processing. The segmentation model SGM segments the first image IMG(1) to identify the terrain features TF(n). The terrain features TF(n) are classified as being traversable or non-traversable. The classification is performed in dependence on the operational capabilities of the vehicle 5. The classification may, for example, reflect the off-road capability of the vehicle 5. A first segmented image IMG(1)(S) corresponding to the first image IMG(1) is shown in Figure 6B. The segmentation model SGM identifies the traversable region TRR of the off-road terrain ORT. Alternatively, or in addition, the segmentation model SGM identifies the non-traversable region(s) UTR of the off-road terrain ORT. As described herein, a route RT1 may optionally be plotted within the bounds of the traversable region TRR identified by the segmentation model SGM. A fourth example of the operation of the segmentation model SGM to identify a traversable region TRR within a section of off-road terrain ORT will now be described with references to Figures 7A and 7B. A first image IMG(1 ) comprising the off-road terrain ORT is captured by the first imaging sensor 21. The first image I MG(1 ) is shown in Figure 7A by way of example. The off-road terrain ORT in this example comprises an un-metalled surface extending between sections of grass. The grass on the left-hand side is labelled as the terrain feature TF(1); and the grass on the right-hand side is labelled as the terrain feature TF(2). The un-metalled surface forms a dirt trail between the sections of grass. There are no formal markings, indicia or street furniture to mark the boundary of the un- metalled surface. The terrain features TR identified by the segmentation model SGM comprise the grass on the left- and right-hand sides of the un-metalled surface. The first image data I MD(1 ) representing the first image IMG(1) is supplied to the segmentation model SGM for processing. The segmentation model SGM segments the first image I MG(1 ) to identify the terrain features TF(n). The terrain features TF(n) are classified as being traversable or non-traversable. The classification is performed in dependence on the operational capabilities of the vehicle 5. The classification may, for example, reflect the off-road capability of the vehicle 5. A first segmented image IMG(1)(S) corresponding to the first image IMG(1) is shown in Figure 7B. The segmentation model SGM identifies the traversable region TRR of the off-road terrain ORT. Alternatively, or in addition, the segmentation model SGM identifies the non-traversable region(s) UTR of the off-road terrain ORT. As described herein, a route RT1 may optionally be plotted within the bounds of the traversable region TRR identified by the segmentation model SGM.

[0079] The segmentation model SGM is trained using a machine learning algorithm (MLA). The training is performed using a computer- implemented method to process a plurality of training data sets. The or each training data set may, for example, comprise image data IMD(n) representing an image IMG comprising at least one terrain feature TF(n). The or each training data set may be annotated to identify the or each terrain feature TF(n). The training data may be annotated to classify the or each terrain feature TF(n) as being traversable by the vehicle 5 and / or non-traversable by the vehicle 5. The or each terrain feature TF(n) may be annotated as being traversable or non-traversable in dependence on the operating capability of the vehicle 5. The training data may be annotated manually, automatically or semi-automatically. The segmentation model SGM can be trained to classify the at least one terrain feature TF(n) in dependence on the operating capabilities of the vehicle 5. For example, the segmentation model SGM may be trained to determine the terrain class TCL of the or each terrain feature TF(n) in dependence on the operating capabilities of the vehicle 5. One or more of the plurality of training data sets may be specific to the operating capabilities of the vehicle 5. For example, one or more of the plurality of training data sets may comprise image data IMD(n) representing an image IMG comprising at least one terrain feature TF(n) determined to be traversable by the vehicle 5; and / or image data IMD(n) representing an image IMG comprising at least one terrain feature TF(n) determined to be non-traversable by the vehicle 5.

[0080] By way of example, a computer-implemented training method 200 is performed by processing a plurality of training data sets. The computer-implemented training method 200 is illustrated in Figure 8. The training data sets comprise a set of first image data IMD(1) representing first images I MG(1 ); and a set of second image data IMD(2) representing second images IMG(2). The training data sets are annotated to provide an indication of the form of the terrain features represented in the first and second images IMG(1), IMG(2). The first image data IMD(1) represents a plurality of the first images IMG(1), wherein the first images I MG(1 ) comprise or consist of a first terrain feature TF(1). The or each first terrain feature TF(1) represented in the first images I MG(1 ) is labelled as being traversable in dependence on the operational capability of the vehicle 5. The second image data IMD(2) represents a plurality of the second images IMG(2), wherein the second images IMG(2) comprise or consist of a second terrain feature TF(2). The or each second terrain feature TF(2) represented in the second images IMG(2) is labelled as being non-traversable in dependence on the operational capability of the vehicle 5. The segmentation model SGM is trained using the plurality of training data sets to differentiate between the first and second terrain features TF(1), TF(2). The resulting segmentation model SGM may be employed to classify terrain features FT as being either traversable or non-traversable in accordance with the method(s) described herein. Alternatively, or in addition, the training data sets may comprise image data IMD(n) representing off-road terrain ORT which does not include any non-traversable terrain features TF(n). Alternatively, or in addition, the training data sets may comprise image data IMD(n) representing a plurality of terrain features TF(n) comprising both traversable terrain features TF(n) and non-traversable terrain features TF(n).

[0081] The system 1 may is operable to classify the or each terrain feature TF(n) in the first image IMG(1) as being traversable or non- traversable. The system 1 may optionally pre-configure one or more vehicle subsystems to facilitate traversal of the one or more terrain feature TF(n). For example, the system 1 may select one of a plurality of subsystem control modes suitable for traversal of a particular terrain feature TF(n). By way of example, the subsystem control modes may include one or more of the following:

[0082] 1. A first subsystem control mode in the form of a comfort subsystem control mode suitable for traversing terrain comprising a paved (metalled) road, motorway or regular roadway.

[0083] 2. A second subsystem control mode in the form of a grass / gravel / snow subsystem control mode (GGS mode) suitable for traversing terrain comprising or consisting of grass, gravel or snow terrain;

[0084] 3. A third subsystem control mode in the form of a mud / ruts subsystem control mode (MR mode) for traversing terrain comprising or consisting of mud and / or rutted terrain;

[0085] 4. A fourth subsystem control mode in the form of a sand subsystem control mode suitable for traversing terrain comprising or consisting of sand (or deep, soft snow);

[0086] 5. A fifth subsystem control mode in the form of a rock subsystem control mode suitable for traversing terrain comprising or consisting of rocky terrain such as a boulder field.

[0087] The subsystem control modes are predefined and the selection may be made automatically or semi-automatically by the system 1. The system 1 may select one of the subsystem control modes to provide appropriate control of the vehicle subsystems. The selection of the subsystem control mode may, for example, be made in dependence on the traversability grade determined for the one or more terrain feature (TF(n).

[0088] It will be appreciated that various changes and modifications can be made to the present invention without departing from the scope of the present application.

Claims

CLAIMS1. A system for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the system comprising one or more processor collectively configured to: receive image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; process the image data using an image segmentation model to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain; wherein, in dependence on an operational capability of the vehicle, each of the one or more identified terrain feature is classified as either being traversable or non-traversable; determine the traversable region of the off-road terrain by identifying a region of the image which excludes any terrain features classified as being non-traversable; and output traversable terrain data representing the traversable region of the off-road terrain.

2. A system as claimed in claim 1 , wherein the region of the image identified as the traversable region of the off-road terrain comprises one or more terrain feature classified as being traversable.

3. A system as claimed in claim 1 or claim 2, wherein the classification of the one or more identified terrain feature comprises determining a surface gradient of the terrain feature, the terrain feature being classified as being a non-traversable terrain feature in dependence on a determination that the surface gradient is greater than a gradient threshold.

4. A system as claimed in any one of claims 1 , 2 or 3, wherein the classification of the one or more identified terrain feature comprises determining a step-change in height associated with the terrain feature, the terrain feature being classified as being a non- traversable terrain feature in dependence on a determination that there is an associated step-change in height which is greater than a step-change height threshold.

5. A system as claimed in any one of claims 1 to 4, wherein the classification of the one or more identified terrain feature comprises determining a height of the terrain feature, the terrain feature being classified as being a non-traversable terrain feature in dependence on a determination that the height is greater than a height threshold.

6. A system as claimed in any one of the preceding claims, wherein the classification of the one or more identified terrain feature comprises allocating one of a plurality of traversability grades to the or each identified terrain feature. A system as claimed in claim 6, wherein the or each identified terrain feature is classified as being traversable if the allocated traversability grade is less than a threshold value; and / or the or each identified terrain feature is classified as being non- traversable if the allocated traversability grade is greater than or equal to a threshold value.

8. A system as claimed in any one of the preceding claims, wherein the image segmentation model is configured to process the image data to classify each of the one or more identified terrain feature as being either traversable or non-traversable.

9. A system as claimed in any one of the preceding claims comprising plotting a route for the vehicle across part or all of the off-road terrain, wherein the route is plotted within the determined traversable region of the off-road terrain.

10. A system as claimed in claim 9, wherein determining the traversable region of the off-road terrain comprises determining a separation distance between a first and a second of the identified terrain features classified as being non-traversable; wherein the determination of the traversable terrain comprises determining that the separation distance is greater than a threshold separation, optionally the classification of the one or more identified terrain feature comprises identifying that the terrain feature comprises or consists of a wheel rut, the terrain feature being classified as being traversable in dependence on a determination that the terrain feature comprises or consists of the wheel rut.

11. A system as claimed in any one of the preceding claims, wherein the classification of the one or more identified terrain feature as being traversable comprises defining one or more traversal direction in which the terrain feature is traversable by the vehicle.

12. A vehicle comprising the system of any preceding claim and at least one imaging sensor for capturing the image data representing an image of the off-road terrain.

13. A method for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the method comprising: receiving image data representing an image of the off-road terrain, the image data being captured by at least one imaging sensor provided on the vehicle; processing the image data using an image segmentation model to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain; classify each of the one or more identified terrain feature is classified as either being traversable or non-traversable in dependence on an operational capability of the vehicle; determining the traversable region of the off-road terrain by identifying a region of the image which excludes any terrain features) classified as being non-traversable; and outputting traversable terrain data representing the traversable region of the off-road terrain.

14. A computer-implemented training method for training an image segmentation model to segment an image to identify terrain features present in an off-road terrain and to classify the identified terrain features in dependence on an operational capability of a vehicle; the method comprising receiving a plurality of training data sets, the training data sets comprising: a first image data representing a plurality of first images, each of the first images representing a first terrain feature, the first image data being labelled as being traversable; a second image data representing a plurality of second images, each of the second images representing a second terrain feature, the second image data being labelled as being non-traversable; and training the image segmentation model to differentiate between terrain features which are traversable and non-traversable.

15. A system for processing image data to identify a traversable region of an off-road terrain, the traversable region representing a region of the off-road terrain which is traversable by a vehicle, the system comprising one or more controller configured to implement an image segmentation model as trained using the method of claim 14 to process image data generated by an imaging sensor, to segment the image into a plurality of image segments, the plurality of image segments identifying one or more terrain feature present in the off-road terrain.

Images