A device and method for road safety

Abstract

A road safety device, comprising: projecting a light pattern 308 on a road; a camera capturing images of the pattern and an overtaking vehicle 304; a classifier model; a processor to analyse distortio

Description

Technical Field of the Invention The invention relates to apparatus and methods for road safety. Specifically a device suitable for determining unsafe overtaking manoeuvres performed on potentially vulnerable road users (for example, bicyclists) and to encourage safer driving. The invention extends to related methods. Background to the Invention Devices for road safety that aid the safety of vulnerable road users, such as bicyclists, include the use of bright and fluorescent clothing. Further, devices known as a “lollipop” “safety flag” or “flash flag”, which is a yellow or orange piece of flexible plastic warning sign arranged to stick out (via a pivoting and typically spring-mounted bracket) from near the rear axle of the bicycle, which encourages drivers to leave more space when overtaking the cyclist may be used. The various visibility aids are effective only by way of ensuring the driver has noticed the cyclist and thus are unreliable at improving safety. US2019256162 describes a device for a vulnerable road cyclist which measures the overtake distance and overtake speed of passing motor vehicles. Different alternative approaches are suggested, including the use of a single ultrasonic sensor, a single pair of infra-red distance sensors arranged at different angles to capture the car passing through their beams sequentially. Ultrasonic sensors can only perform measurements at a comparatively low rate and will tend to experience interference from similar nearby devices. The difficulty with infra-red sensors is that curved semi-reflective vehicle body panels can cause anglebased infrared distance sensors to produce less accurate data, whilst direct sunlight and polished black vehicle paint can disrupt the operation of most of the more affordable time-of-flight based infrared distance sensors, and in general their very extremely narrow fields of view, coupled with sway / motion of a bicycle or other vulnerable road user vehicle can cause errors when determining the time of entry and departure of the beam, especially with cars that have a long front bonnet. One difficulty with the idea of using stereoscopic cameras is the overall cost of not only the additional camera, but a processor board capable of frame-synched control of two cameras, the need for greater processing power (or internet bandwidth and cloud processing), and the need for higher quality components due to difficulties in determining distances to curved and semi-reflective car panels. In that prior art document, along with other alternatives for how to measure the distance of the passing vehicle, there is also a brief mention of using a single video camera that measures the distance to a passing vehicle based on an estimate of that vehicle’s size, or identifies the location of a passing vehicle wheel with the camera orientation being accurately calibrated using ‘standard street furniture’ (which perhaps can be objectively inferred as primarily meaning road signage with nationally standardised geometry). In this regard there is no suggestion that any additional measurements or sensor components are required. Vulnerable road users are at risk of dangerous manoeuvres performed by road vehicles such as cars, particularly overtaking manoeuvres. Providing a solution to help improve road safety would improve the safety of vulnerable road users such as cyclists. Summary of the Invention According to a first aspect, the invention provides a device for road safety, comprising: a projector configured to project in use a pattern of light on a section of a road; a camera configured to capture in use one or more images of the pattern of light projected on the section of road and an overtaking vehicle, the camera and projector having a predetermined difference in viewing direction of the section of road; a classifier module comprising a model that is trained to classify one or more features of a vehicle using the one or more captured image of the overtaking vehicle; a computer processor configured to receive the one or more images, and to analyse distortions of the pattern of light, and further configured to: - map, based on the distortions of the pattern of light, a surface of the section of the road; - classify, using the classifier module, one or more features of the overtaking vehicle; - determine, based on the map, the classified feature of the overtaking vehicle and the predetermined difference in viewing direction, a relative location between the device and the overtaking vehicle. The device for road safety has a projector that is used to project a known pattern of light with onto a surface of a road. The known pattern of light deforms according to the shape and contours of the surface (for example a road surface). The camera is positioned a known distance from the projector and captures one or more images of the deformed known pattern. This is a technique known as structured light. Multiple images can also be captured in a video format. The computer processor processes the one or more images and calculates the co-ordinates of points on the surface using triangulation principles. For example, a triangle is formed between the projector, the camera and a point on the surface. The computer processor knows the angles and distances within this triangle and can therefore determine locations of the surface in three-dimensional space. The computer processor can therefore map a surface of a road using only a projector and a single camera. This type of mapping is beneficial because of its accuracy and efficiency. Detailed and accurate mapping can be performed by processing information from only a projector and single camera. A model can be trained to classify a feature of a vehicle. One example of such a model is an algorithm programmed to detect features of a particular shape, for example an algorithm selected and programmed via testing and improvement so as to detect ovals or circles corresponding to wheels or tyres of motorvehicles would then duly detect such wheels or tyres in images. Algorithms for detecting predetermined shapes are well known in the art and can be readily designed by a person skilled in software programming. As another example, the model is trained using a dataset of labelled images of multiple vehicles and a particular feature of a vehicle (for example, a tyre or more specifically, a front or rear nearside tyre). As is known in the art, these images can be used to train a machine learning model to automatically detect those features when presented with new images (for example, the images captured by the camera of the device in use). For example, the labelled images may go through a feature extraction step during a training phase and the model learns to associate features and patterns in the images with the corresponding labels. The model can be tested and evaluated using known and labelled images once it has been trained. Once trained, the model can input new and unseen images and predict a likely label based on the learned patterns. A known technique for image classification is to use a Convolutional Neural Network (CNN). The model can be trained on the device. Alternatively, the model is trained externally of the device and transferred to the device prior to operation. The model may be continually updated with new images. It is also envisaged that the device can transmit captured images to an external for processing through the model. The results can then be transmitted to the device or a different external device (for example, a mobile phone). Using the mapped surface and features of a vehicle captured from a camera, the relative location of the device which has the camera and the vehicle can be determined. Advantageously, the processor can determine the distance between the device and an overtaking vehicle, for example the distance in a specific direction such as lateral to an in-use direction of movement of the device. This allows the device to automatically or the user to manually determine whether the overtaking manoeuvre was dangerous. For example, if the overtaking vehicle passed within threshold distance (for example 1.5 metres) of the device, the overtaking manoeuvre is deemed to be dangerous. The user of the device can then use the captured image or images to provide evidence of the dangerous manoeuvre. Generally, the device is for mounting to bicycle and is for monitoring other vehicles (for example, cars and vans). Thus, the terms “vehicle” and “overtaking vehicle” refer to vehicles that are remote from the device and the object that the device is mounted on respectively. Whilst, the device is primarily described with reference to bicycles and cyclists, the device could be fixed other objects and vehicles (even motorbikes, cars and other motorvehicles, although it would typically be used by a vulnerable road user, particularly for bicycles, mopeds, e-scooters, mobility scooters, horses / equestrians and tricycles). Generally, the invention is described with reference to an overtaking vehicle but it is also envisaged that the device could be used with reference to an approaching vehicle that does not perform an overtaking manoeuvre. In some embodiments, the device for road safety is for use on a user’s road vehicle having an axis of travel, and in an in-use configuration, the relative location is determined between the device and the overtaking vehicle, is a lateral distance with respect to that axis of travel. For example, the user’s road vehicle is a preferably a bicycle but could be a vehicle of other vulnerable road users (for example, a scooter or a moped). Generally, the device captures overtaking manoeuvres where the users road vehicle is traveling in a first direction along an axis of travel (for example, along a road) and the overtaking manoeuvre occurs in the same direction but a lateral distance away from the road user’s axis of travel. In some embodiments, the projector is configured to project light in the infrared spectrum or near-infrared spectrum. The infrared spectrum refers to light that has a wavelength that ranges from around 750nm to 1mm. The near-infrared spectrum is a sub-range of the infrared spectrum and refers to light with a wavelength of 750nm to 1400nm. Advantageously, the infrared and near-infrared spectrum are not visible to the naked eye. Cyclists and other road users may not wish project visible light patterns onto the road surface. Therefore, using light that is not visible to the naked eye may improve adherence in using the road safety device. Additionally, infrared and near-infrared light sources are safe and cost effective compared to other invisible light sources. In some embodiments, the pattern of light is generated by directing a laser or LED through an optical element. A single light source such as a laser or LED can be used to generate a pattern of light by shining the light through an optical element (for example, a diffraction grating or diffractive optical element). The optical element may have pattern of varying structure that produces a known pattern of light on a surface when light from an LED or Laser is shone through the optical element. Advantageously, only a single light source is required to generate the known pattern. This reduces the complexity and cost of the device. In some embodiments, the pattern of light is a two-dimensional pattern and / or the two-dimensional pattern is a line pattern. The pattern of light that is projected from the device can generate any two-dimensional pattern on the surface. For example, the pattern can be a dot pattern or a line pattern or a combination of thereof. The patterns of light that is chosen is dependent on the qualities of the camera and the properties of the surface that the pattern of light is projected on to. Another consideration is ease of manufacture particularly if a single light source with an optical element is used. In some embodiments, a line pattern is generated. For example, a pattern of parallel lines is used as the pattern of light. Parallel lines may be beneficial because of ease of detection compared to other patterns. An array of dots may be beneficial compared to an array of lines, as it can provide a stronger signal without requiring a stronger laser. In some embodiments, the camera is oriented to capture a front or a rear nearside tyre of the overtaking vehicle. The camera is oriented and positioned to take one or more images of the nearside of an overtaking vehicle. The nearside of the vehicle is the side of the car that is closest to the kerb, or roadside, when travelling in a direction along an axis that is the same as direction but laterally apart from the axis of travel of the device, and being an overtaking vehicle its nearside is also nearest to (and typically therefore visible from), the device. The camera is orientated and positioned such that the front and / or rear nearside tyres of an overtaking vehicle are typically captured in use. Typically, the camera is oriented to capture a vehicle number plate of the overtaking vehicle, typically at least the front number plate (if present). The camera (or an additional camera) may also be oriented and positioned to take one or more images of the rear vehicle number plate. The number plate, license plate and registration plate are synonymous. Capturing the number plate allows the user of the device to identify the overtaking vehicle as well as find out information of the overtaking vehicle (for example, the vehicle tax and MOT status). The vehicle information can then be used as evidence to identify a vehicle that performed an unsafe or potentially unsafe overtaking manoeuvre. The vehicle information can be shared with authorities such as the police and other relevant entities (for example, motorvehicle insurance companies). It is to be understood that a single camera can be oriented to capture a front or a rear nearside tyre of the overtaking vehicle and a vehicle number plate of the overtaking vehicle. However, it may be beneficial to provide a dedicated camera to capture the number plate. In some embodiments, the model is a trained to classify a front or a rear nearside tyre of a vehicle. In some embodiments, an automated number plate recognition system is provided. An automated number plate recognition (ANPR) involves a computer processor arranged to read a number plate of a vehicle, either in real time or later on request. ANPR systems are used ubiquitously in road safety, road and parking management and other related industries and are well understood in the art. Use of an ANPR system allows for automatic rapid determination and identification of a vehicle. The automated number plate recognition (ANPR) system may be on the device or external to the device. If the ANPR is external to the device then the number plate information is transmitted to the ANPR system and the results are transmitted back to the device or to another external device (e.g. a mobile phone). In some embodiments, the relative location between the device and the one or more features of the overtaking vehicle is determined by: - determining, using the captured one or more images of the front or rear nearside tyre and the map of the surface of the road, a notional contact position that lies at, or proximately inboard of, the outboard extremity of that tyre’s outer wall; and - determining the relative location between the notional contact position and the device. Due to the geometry of a tyre it is difficult (perhaps impossible) to precisely determine a near-side position in which the tyre contacts the road using images captured from a camera. To overcome this problem, when trying to determine the tyres position on the road using the previously discussed mapped surface, a determination is made for a notional contact position (i.e. a notional point at which the tyre contacts the road). This point is at or proximately inboard of the outboard extremity of that tyre’s outer wall. This is discussed in more detail below with reference to figures. The relative location and / or distance between the notional contact point and the device is then determined. Typically, the lateral distance between the device and the notional contact point is greater than the lateral distance between the device and the near side wall of the tyre but preferably no more than 10cm greater, more preferably no more than 5cm greater. As such, by measuring the lateral distance of a notional contact point, it is possible to identify a lateral distance of the overtaking vehicle with a risk only of slightly overestimating that distance, and no risk of underestimating it. The advantage of this is that the assessment is reliable, whilst any imprecision is achieved in a manner that is fair to the driver of the overtaking vehicle. The present invention determines the apparent contact point of the passing vehicle’s tyre with that section of road that is most proximate in the measurement direction of interest, such as the direction lateral to the direction of travel of the vulnerable road user, and then determines the distance of that apparent contact point in that direction based on the shape, e.g. a road topology map, based on the measurement of the section of the road). It has been identified that with a freshly valeted jet-black and semi-reflective vehicle panel, the camera may not be able to detect the structured light pattern in direct sunlight (without increasing the intensity of the structured light / LED / laser pattern to either require unreasonable power consumption or risking being unsafe), with the result that a minority of vehicles might essentially be exempt from detection. This would not only limit success in reporting dangerous overtakes, but it creates the risk that the drivers of such vehicles might decide that they are not being monitored, and the deterrent would be ineffective on them. For this reason the described approach, even an indirect measurement, or one requiring greater processing power is advantageous. In some embodiments, the notional contact position is a point on the periphery of the tyre that is nearest the mapped road surface from the viewpoint of the camera. In other words, there is a point at the bottom of a tyre that is almost in contact with the road surface that can be determined using images from a camera. The relative location and / or distance between the notional contact point and the device is then determined. In some embodiments, the camera is configured to capture in use at least two images of the overtaking vehicle and the computer processer is configured to calculate a relative velocity between the device and the overtaking vehicle; wherein the relative velocity is calculated by determining the relative locations of the overtaking vehicle and the device over a period of time. Multiple images can be captured of the overtaking vehicle and the relative distance of the overtaking vehicle and the device can be determined as discussed above. Additionally, the multiple images can have a time associated with them. The time that each image is taken and the location of the device relative to the overtaking vehicle can be used to determine the relative velocity of the overtaking vehicle in relation to the device. If a relative velocity is above threshold value (e.g. 70 mph) then it may be determined that an unsafe manoeuvre was performed, but in practice an unsafe manoeuvre is usually assessed based on a combination of velocity and distance (e.g. a relative speed over 30mph may be assessed as dangerous if the lateral passing distance was less than 1,5m). In some embodiments, the computer processor is configured to receive the absolute travelling velocity of the device and to calculate, using the absolute velocity of the device and the calculated relative velocity, the absolute velocity of the overtaking vehicle. The absolute velocity of the overtaking vehicle can be determined most easily if the absolute velocity of the device is known. In embodiments, the computer processor receives a measure of the absolute velocity of the device. The absolute velocity of the device can be determined by means of a GNSS receiver, or by measuring the wheel speed of the vehicle to which the device is attached, or by a combination of GNSS receiver and an accelerometer (the two measurements being combined to provide real time forward velocity, since GNSS can typically only provide time-averaged velocity), or by a combination of GNSS receiver and measuring the wheel speed of a vehicle (for example a bicycle) carrying the device, or by any other means. The absolute velocity of the device is the used to calculate the absolute velocity of the overtaking vehicle. Advantageously, the absolute velocity of the overtaking vehicle can be determined. If the absolute velocity of the overtaking is above a threshold (for example, 30 mph) and the distance between the device and the overtaking is below a threshold distance (for example, 1.5m), a determination is made that the overtake manoeuvre was unsafe. This information can be used as evidence and provided to authorities and / or other relevant entities (e.g. insurance companies). Determining the absolute rather than relative velocity of the vehicle is very useful, since if the device (E.g. on a bicycle) is itself travelling at 28mph, then a close overtake with a relative speed of 2mph (i.e. an absolute speed of 30pmh) is very dangerous, whereas if the device / bicycle is stationary then a close overtake with a relative speed of 2mph is typically very safe. As such the best determination of whether an overtake was dangerous takes into account the absolute speed of the overtaking vehicle, the proximity of the overtake. The determination of whether an overtake was dangerous may also optionally benefit from including further factors such as the device / bicycle speed, the size / class of overtaking vehicle (e.g. motorbike vs heavy goods vehicle), lighting conditions (day vs night), weather (calm vs storm) etc. In some embodiments, the device further comprises a light source that is distinct from the projector, configured to substantially evenly illuminate at least one part the section of the road. Shining a light on the road (in addition to the light produced by the projector) makes determining the feature of the overtaking vehicle easier. This is particularly advantageous when it is dark outside. Preferably, the light source is red, thereby providing a rear light for the vehicle that the device is mounted on - e.g. a rear bicycle light. In some embodiments, the light source is configured to illuminate at least part of the section of the road corresponding in use to a direction where a vehicle tyre is to be measured, with a greater radiant light intensity compared to the radiant illumination another part of the section of road corresponding in use to a direction where a number plate of the overtaking vehicle is to be captured. When a light source is used to illuminate the road surface, that same light can cause reflections off of the number plate that can saturate the camera sensor, and obscure the detail and make it difficult or impossible to read. Therefore, it can be beneficial to avoid strong or high levels of illumination where the number plate is expected to be in order to allow for accurate determination of the relative location of the overtaking vehicle without causing the number plate to be overly saturated by the illumination. In some embodiments, the device further comprises at least one additional sensor for independently measuring the relative location, or the relative location and vehicle velocity, wherein the at least one of such additional sensor comprises an infra-red, ultrasonic, or radar distance sensor. It is well understood in the art that sensors exist that can determine an overtaking vehicles relative or absolute location along with the overtaking vehicles relative or absolute velocity. By providing additional and alternative sensors to detect the relative location and the vehicle velocity, there is a second source of information to corroborate findings from the camera and projector based determination. Additionally, the alternative sensors may provide a quicker analysis of the relative and / or absolute location or the relative and / or absolute speeds of the overtaking vehicle. The camera and projector based determination can then be used to provide video evidence. This evidence can be provided, for example, to the authorities or other entities (for example, insurance companies). In some embodiments, the device further comprises a red light oriented to the rear in an in-use position on a bicycle. The red light orientated and positioned at the rear of the bicycle (or other vehicle of a vulnerable road user) has the effect of a rear bicycle light. This allows the bicyclist to notify their position on the road in the usual and well known way. The red light may be constant or flashing to increase visibility. In some embodiments, the red light is in the form of a symbol or an image with time-varying shape or motion, such as to present a sign to approaching motorists. The red light can operate in a way to distinctively indicate to other road users that the bicycle is fitted with the device. That is to say, the red light provides a sign. Other road users wishing to overtake the bicycle may be encouraged to only do so safely, and the further objective of influencing safer driving would be achieved. In some embodiments, a connector for attachment to a bicycle. In some embodiments, the connector is configured to attach to the seat post or seat of the bicycle. Generally, the device comprises a bracket for connection to a bicycle saddle. Typically, the bracket is arranged to lock onto both of a pair of standardised seat-post-connection bars of a bicycle saddle undercarriage, such that the connection between the bracket and the seat-post-connection bars governs the orientation of the device relative to the saddle. This has the advantage that, because a bicycle saddle is generally very well aligned with the bike frame, the camera and projector (and any other sensors) will also have a fixed orientation relative to the bicycle, with the result that the camera, projector and any other sensors can be configured so that they will point in appropriate, consistent, directions relative to the bicycle, which enables more accurate mapping and determination of overtaking vehicle distance (and velocity / speed if measured) and ensures that the camera can be arranged so that it will collect images including the front and rear nearside of the overtaking vehicle and the overtaking event as a whole. Helpfully, the bracket may have a press-release connection to a device housing, such that upon the press-release connection being pressed, the device housing can be removed from the bracket and from the bicycle. This has the advantage of ensuring that the device can be easily removed either for charging a battery of the device, or to avoid it being stolen when the bicycle is left outside unattended. Optionally the press release connection includes a data connection such that when the housing is fitted to the press-release connection the data connection is connected. Generally, the determination made by the device with regards to relative and or absolute velocity of an overtaking vehicle and relative distance between the device and the overtaking vehicle is output to a remote device. For example, the remote device may be a server or a mobile device. The mobile device preferably belongs to the user of the device. In some embodiments, the device comprises a transceiver configured to transmit data to a remote device. The data can be transmitted in any known way (for example, over Wi-Fi networks or mobile networks). In some embodiments, the device comprises an output module (for example, a screen) configured to display data, According to a second aspect, the invention provides a method for road safety, comprising the steps of: - projecting a pattern of light on a section of a road; - capturing one or more images of the pattern of light projected on the section of road and an overtaking vehicle, the camera and projector having a predetermined difference in viewing direction of the section of road; - analysing distortions in the pattern of light in the captured one or more images; - mapping a surface of the section of the road based on the distortions of the pattern of light; - classifying one or more features of the overtaking vehicle; - determining a relative location between the device and the overtaking vehicle based on the map, the classified feature of the overtaking vehicle and the predetermined difference in viewing direction. In some embodiments, the camera is orientated to capture a front or a rear nearside tyre of the overtaking vehicle. In some embodiments the method comprises: - determining, using the captured one or more images of the front or rear nearside tyre and the map of the surface of the road, a notional contact position that lies at, or proximately inboard of, the outboard extremity of that tyre’s outer wall; and - determining the relative location between the notional contact position and the device. In some embodiments, the method comprises capturing in use at least two images of the overtaking vehicle and calculating a relative velocity between the device and the overtaking vehicle; wherein the relative velocity is calculated by determining the relative locations of the overtaking vehicle and the device over a period of time. In general it is envisaged that the device uses structured light and a camera to measure the road (or land) surface, and then uses a camera to locate part of a remote vehicle which is known to remain in contact with the road surface, and to map it to the measured road surface, thereby determining the location of that part of the vehicle, and for example thereby determining an upper bound on the distance of that vehicle in at least one direction. The device may alternatively or additionally use other methods to measure the location of the vehicle, such as using the structured light and camera, or using stereo cameras. As an alternative to using a camera to measure the distance and velocity of the vehicle, the device may instead use a distance sensor such as one or more infrared or ultrasonic sensors to measure the distance and velocity of the vehicle. However this application primarily envisages the former approach of structured light to primarily map the road surface, as a precursor to mapping the location of an identifiable ground-contact vehicle part such as a lower extremity of a tyre in contact with the road. Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, it is preferable that the device or system comprises an absolute device motion sensor arranged to measure its own absolute velocity or speed with respect to a road or other path on land. This may be implemented as a GNSS sensor, or a road wheel sensor (which may be provided as a wirelessly connected but physically separate component), or as a one-dimensional or two-dimensional optical flow sensor oriented to measure the road surface (which may be implemented via the main or auxiliary camera, and may be measured via on-board, or remote, or cloud, processing). The advantage is that this enables measuring the absolute velocity / speed of the remote vehicle, rather than only its relative velocity / speed. This is especially useful when assessing or measuring how dangerous an overtake by a remote vehicle was, from a moving platform such as a vulnerable road user’s vehicle such as a bicycle. If for example a car overtakes a bicycle with a relative speed of 1mph and with an extremely narrow passing distance, this may be considered quite safe if the bicycle is stationary, yet extremely dangerous if the bicycle is travelling at 30mph. Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, preferably the absolute device motion sensor also comprises a secondary / second device motion sensor, to either calibrate, check the reliability of, and / or improve the time-precision of a primary / main / first sensor. For example if the primary absolute device motion sensor is a GNSS position sensor, then a secondary device motion sensor may be an accelerometer (at least in the axis of travel) and optionally also a gyrometer (at least measuring incline of travel to adjust the accelerometer output for any hill / slope). If the primary absolute device motion sensor is a wheel sensor (e.g. to measure revolutions of the road-contact wheel, as is used on bicycles), the secondary device motion sensor may be a GNSS sensor, or an accelerometer (at least in axis of travel) optionally with gyrometer (to again account for road incline), or a one dimensional or two dimensional optical flow sensor directed to measure relative road motion (albeit potentially only at slower speeds). Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, more generally the relative motion of the road (or path / land) itself relative to the device / sensor is preferably measured within the laser / structured light projection area / field of view. This can be implemented most easily via a technique commonly known as ‘optical flow’ measurement, which is a computer vision technique that simply measures motion of any observable features, such as the surface texture of the road / tarmac / gravel or any paint markings etc. Optical flow is a well-known technique in for example drone control sensors, and can be implemented using a COTS optical flow sensor with built in computer processor. Numerous suitable algorithms have been developed and described in the art for two dimensional optical flow measurement, such as phase correlation, block-based methods, differential methods, and discrete optimisation methods, and several are available in COTS products (often using the PMW3901 or PAW3902 processor chips and various lenses) or in open source software packages such as OpenCV. One dimensional optical flow is an easier problem to solve, and can most simply be done using phase correlation, however fewer COTS sensors are available for this. One difficulty with two-dimensional optical flow sensors is that most long-range COTS sensors are designed to look essentially downwards from a flying drone, and thus often have a wide and nearly square field of view in the region of 40-50 x 40-50 degrees. For road vehicle use it is preferable to have a smaller field of view (e.g. with at least one side being <20 degrees or preferably <10 degrees) oriented to measure sideways and / or forwards / backwards from a bicycle or other road vehicle with respect to its in-use direction of travel. It is also preferable to have a higher aspect ratio to the rectangular field of view, such as at least 3:1 or at least 10:1, since a wide component in the direction of in-use travel aids measurement of fast moving road surface, whilst a narrow component in the lateral direction (with respect to in-use direction of the vehicle it is mounted to) reduces the problem that when looking sideways the optical flow rate varies sharply across the field of view (possibly with further variation caused by any sway / tilt of the moving vehicle that the sensor is attached to, e.g. a bicycle). Given this potential need for measurement of a narrower or more specific field of view, it may be preferable to perform optical flow measurement in the relevant region of the field of view of the camera video feed using a computer processor. Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, by determining the rate and direction of optical flow (E.g. in radians / second) and the distance to the relevant section of road being measured, it is possible to determine the velocity of the road relative to the sensor, and then the absolute velocity of the remote vehicle can be determined by adding the relative motion of the vehicle part to the relative motion of the road surface. Two-dimensional optical flow sensors which are capable of measuring at distances of greater than 30cm (such as those for drones rather than in optical mice where high illumination is available) have a maximum optical flow rate they can measure of typically between 7.2 and 9.5 radians per second and certainly under 10 radians per second. Especially in the case that there is no secondary device motion sensor, it is preferable that the optical flow sensor can measure up to at least 20 radians per second, preferably at least 35 radians per second, although this is a demanding requirement, and aside from the faster processor requirement, a wider and more expensive optical aperture will typically be required, especially for use in low light / illumination conditions. That said, if the optical flow is only measured in one dimension this may be more easily achievable at lower cost. Therefore more generally, the device comprises a one-dimensional optical flow sensor, oriented to measure the absolute speed or velocity of the device with respect to a road or other land surface, and with its axis of measurement aligned with an in-use direction of a vehicle it is for use with or mounted on, and preferably it is arranged to measure at least 20 radians per second, more preferably at least 35 radians per second. Preferably this data is then used to calculate the absolute velocity or speed of a remote vehicle based on its measured relative velocity, for example to measure the absolute velocity or speed of a passing road vehicle. Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, it is beneficial to provide two different methods of determining the location / and or velocity (preferably both) of the remote vehicle, including a first method arranged to provide immediate results (e.g. involving distance sensor(s) based on ultrasonic, infrared (e.g. lidar) and / or radar operation), and a second method providing provable results (involving either stereoscopic cameras or structured light camera data). The accuracy of the first and second methods need not be the same. This has the advantage that the user can be readily provided with immediate results to enable him / her to decide whether the incident merits reporting to authorities, whilst still ensuring that a submission to the authorities and later court action is backed up by provable data. Irrespective how the device measures the location and / or velocity of a remote vehicle or what features it has to do so, it is beneficial to provide a radar sensor oriented in use to measure distance to the overtaking vehicle. This has an advantage in size, weight and potentially measurement accuracy. Preferably the radar sensor is a pulsed radar provided with distance and (radial) velocity measurement output. An example of such a radar is the Acconeer (TM) A121 (TM) radar sensor. Preferably the radar is used to measure both lateral overtake distance, but also to estimate or determine the approach or departure velocity, based on the distance and variation in radial velocity determined with respect to time, as the overtaking vehicle approaches and departs. Optionally such a radar sensor is provided with a radar lens to reduce its field of view, and increase signal to noise from the overtaking vehicle. Such lenses or lens designs are available for the A121 radar chip albeit that these are circular convex lenses that reduce the horizontal and vertical fields of view by the same amount. Preferably the radar lens provides a narrower field of view vertically than horizontally. This has the advantage of increasing signal to noise ratio from the overtaking vehicle (as opposed to the road surface) whilst still capturing approach velocity data from the vehicle as it approaches. Such a radar lens can be achieved using a convex lens that has greater convex curvature wrt the vertical plane, than the horizontal plane in use (or may have no curvature or be concave in the horizontal plane). Various materials can be used that are transmissive to radar in the relevant frequency band, and a common material is PLA (Polylactic acid) plastic, which has the benefit of suiting construction via extruder based 3D printer. For example a cylindrical lens can be used having a height to width ratio of 1:1 to 4:1 preferably substantially 2:1 (as an example of a 2:1 ratio lens is a cylindrical lens that in cross section is substantially a semi-circle, although a more symmetric convex lens is preferable). The radar lens preferably has crenulations or corrugations, comprising surface features that are smaller than the wavelength of radar used, such as to cause smoother or stepwise variation in refractive index. This has the benefit of reducing reflections from the front and / or rear surfaces of the lens. Any feature in one aspect of the invention may be applied to any other aspects of the invention, in any appropriate combination. In particular device aspects may be applied to method or use aspects and vice versa. In all aspects, the invention may comprise, consist essentially of, or consist of any feature or combination of features. Brief Description of the Drawings The invention will now be described, purely by way of example, with reference to the accompanying drawings, in which; Figure 1 is a block diagram of a device according to an embodiment of the invention; Figure 2 is a flow diagram of a method for road safety. Figure 3A is a perspective illustration an in-use scenario of a device according to an embodiment. Figure 3B is top-view illustration of an in-use scenario of a device according to an embodiment. Figure 3C is top-view illustration of an in-use scenario of a device according to an embodiment. Figure 4 is an illustration of the device in use according to an embodiment. Figure 5 is a schematic view of a tyre and the view from a camera according to an embodiment. The drawings are for illustrative purposes only and are not to scale. Detailed Description Figure 1 is a block diagram 100 of a device 101 according to an embodiment of the invention. The device 101 comprises a projector 102, a camera 104, a classifier module 106, a memory module 107, a computer processor 108, a light source 110, an additional sensor 112, a red light 114, a connector 116, a battery 120, an automatic number-plate recognition (ANPR) system 124 and a transceiver 126. The transceiver 126 is in connection with an external output module 122. The connector 116 is connected to a bicycle 118. The projector 102 projects a pattern of light onto a surface of a road. Although any suitable pattern is envisaged, the pattern shown in the figures is a parallel line pattern. However, alternative patterns are also envisaged. The parallel lines are projected onto a road surface 306 (shown in figure 3A) and deform according to the shape and contours of the road. The projector 102 projects light in the infrared spectrum so that the projections are not visible to the naked eye. In the example shown in the figures, the projector 102 produces the pattern of light using a single light source 110 (such as an LED) with a diffractive optical element (shown in Fig. 4). The camera 104 captures multiple images of the road surface 306 and, therefore, captures the distortions (the distortions are shown in figures 3B and 3C). The distance between the camera 104 and the projector 102 is fixed at a predetermined distance, say 10cm. The computer processor 108 determines position of points on the surface of the road using triangulation the triangle in the triangulation is formed the projector 102, the camera 104 and the road surface 306. Therefore, the computer processor 108 can understand distances and angles within the triangle and therefore can determine where each point is in space in relation to the device 101. Therefore, the computer processor 108 maps a surface of the road. The camera 104 is orientated to capture the road surface 306 as mentioned. The camera 104 is also oriented to capture the front or rear nearside tyre of an overtaking vehicle 304 (shown in Figure 3C and 5).The camera 104 is also orientated to capture the a vehicle number plate of the overtaking vehicle 304. Whilst any suitable algorithm can be used to identify and locate the vehicle tyre (and indeed the lower near-side periphery of the tyre), and it need not be an CNN based algorithm. If it is a CNN based algorithm then the classifier module 106 includes a machine learning model that has been trained to identify images of a front nearside tyre of a vehicle. The model has been trained using a Convolutional Neural Network (CNN) in a known way. In more detail, the machine-learning model is trained using labelled images of the front nearside tyre of different vehicles captured in the real world. The labelled images are pre-processed and prepared for the model. For example, the labelled images are re-sized accordingly. A CNN is then used to extract features and patterns of features from labelled images. In this way, the model learns to associate features and patterns in the images with the corresponding labels. The model is tested using known images. Once trained and tested, the model is able to predict whether an image includes a front nearside tyre with a high degree of accuracy. Once trained, the model is transferred to the device 101 for use. Alternatively, the images captured the camera 104 are transmitted to an external device, with includes the model, for processing and the results are transmitted back to the device 101. Using the mapped surface of the road and the identification of the tyre, the computer processor 108 can determine the relative location of the device 101 the vehicle (shown in figure 3A and 3C). In more detail (and as shown in figures 3A-3C), the computer processor 108 can determine the relative distance between the device 101 and an overtaking vehicle 304, for example the distance in a specific direction such as lateral to an in-use direction of movement of the device 101. The computer processor 108 can also determine whether the overtaking vehicle 304 is within a threshold distance of the device 101 (for example, 1.5 metres). Although this example is described with reference to an overtaking manoeuvre, it could also apply to a close following or “tailgating” manoeuvre. In this way, the computer processor 108 can automatically determine whether the overtaking vehicle 304 was dangerously close to the device 101 and, by extension, the vehicle to which the device 101 is attached. If the computer processor 108 determines that a vehicle is within a threshold distance, the computer processor 108 may save images taken from the camera 104 of the vehicle and over taking manoeuvre and store them in the memory module 107. The relative location between the device 101 and the tyre is determined by first determining a notional contact position 504 that lies at, or approximately inboard of, the outboard extremity of the tyres outer wall. Using a camera 104 and projector 102 is difficult to determine the actual contact position where a tyre contacts the road surface 306, this is due to the geometry of a standard tyre. As shown in figure 5 and described below, the notional contact position 504 is a position that is approximates the actual contact position. The computer processor 108 determines the relative location or distance between the notional contact position 504 and the device 101. Importantly, using the notional position as described above, by measuring the lateral distance of a notional contact position 504, it is possible to identify a lateral distance of the overtaking vehicle 304 with a risk only of slightly overestimating that distance, and no risk of underestimating it. The advantage of this is that the assessment is reliable, whilst any imprecision is achieved in a manner that is fair to the driver of the overtaking vehicle 304. In this way, the computer processor 108 can determine the relative location of the bicycle 118 that the device 101 is mounted on and the overtaking vehicle 304. The camera 104 captures multiple images or video sequences. The computer processor 108 then calculates a relative velocity between the device 101 and the overtaking vehicle 304; wherein the relative velocity is calculated by determining the relative locations of the overtaking vehicle 304 and the device 101 over a period of time. The period of time is the time between images or time between frames in a video. If a relative velocity is above threshold value (e.g. 70 mph) then it may be determined that an unsafe manoeuvre was performed, but in practice an unsafe manoeuvre is usually assessed based on a combination of velocity and distance (e.g. a relative speed over 30mph may be assessed as dangerous if the lateral passing distance was less than 1,5m). In this way, the computer processor 108 calculates the relative velocity of the bicycle 118 that the device 101 is mounted on. The computer processor 108 can also determine the absolute velocity of the overtaking vehicle 304 by understanding the absolute velocity of the device 101 and therefore the absolute speed or velocity of the bicycle 118 that the device 101 is mounted on. The absolute velocity of the device 101 can be determined in a number of ways using an additional sensor 112. The additional sensor 112 can be GNSS receiver. The GNSS receiver can determine the time-averaged velocity of the device 101. Once the time-averaged velocity is known, the absolute speed of the overtaking vehicle 304 can be calculated by addition in a manner known in the art. The additional sensor 112 can be any sensor or number of sensors that calculate the absolute speed and or velocity of the device 101. The additional sensor 112 can also be a sensor used to provide information to corroborate findings from the camera 104 and projector 102. The additional sensor 112 is a distance sensor for independently measuring the relative location, or the relative location and vehicle velocity. For example, an infra-red, ultrasonic, or radar distance sensor. It is well understood in the art that sensors exist that can determine an overtaking vehicles relative or absolute location along with the overtaking vehicles relative or absolute velocity. The sensor 112 is shown as being stored on the device 101. Although it is also envisaged that the sensor 112 be external to the device 101. The sensor 112 can determine the absolute speed or velocity of the device 101 and can then send the results to the device 101 via the transceiver 126. The camera 104 is orientated to capture an image of the vehicle number plate. The image of the vehicle number plate is then input into an automated vehicle number plate recognition (ANPR) system 124. The ANPR system 124 takes the image and can automatically detect the number plate of the vehicle. ANPR systems are well known in the art. The ANPR system 124 is shown as being stored on the device 101. Although it is also envisaged that the ANPR system 124 be external to the device 101. In this way, the image of the vehicle number plate is transmitted through the transceiver 126 to an external ANPR system 124. The external ANPR system 124 can then send the results back and received by the device 101 via the transceiver 126. The computer processor 108 then stores the vehicle number plate of the overtaking vehicle 304 with the captured images of the dangerous manoeuvre in the memory module 107. The computer processor 108 can output stored information from the memory module 107 to an output module 122. In this way, the computer processor 108 can output images and / or videos of a dangerous manoeuvre performed by the overtaking vehicle 304 along with the number plate information of the overtaking vehicle 304. The output module 122 may be a display screen on the device 101. Alternatively or additionally, the output module 122 may be external to the device 101. In this way, the transceiver 126 will transmit the outputted images and / or videos of the dangerous manoeuvre performed by the overtaking vehicle 304 along with the number plate information of the overtaking vehicle 304 to an external device. Operating the device 101 in dark environment (for example, at night) may pose a particular challenge. To overcome this challenge, the device 101 includes an additional light source 110 that is distinct from the projector 102. The light source 110 emits visible light and is used to illuminate the surface of the road. Shining a light on the road (in addition to the light produced by the projector 102) makes determining the feature of the overtaking vehicle 304 easier. This is particularly advantageous when it is dark outside. The light source 110 illuminates the road section where a tyre is to be expected to a greater extent or with a greater radiant light intensity compared to where a number plate is to be expected. This allows the camera 104 to capture the tyre without oversaturating and therefor obscuring the number plate. Therefore, the computer processor 108 uses clearer captured images to make determinations such as relative location. The clearer images are also used by the ANPR system. The light source 110 may be red, thereby also providing a rear light for the bicycle 118 that the device 101 is mounted. The device 101 also comprises a red light 114 that is orientated and positioned at the rear of the bicycle 118 has the effect of a rear bicycle light. This allows the bicyclist 302 to notify their position on the road in the usual and well known way. The red light 114 may be constant or flashing to increase visibility. The connector 116 is attached to a seat of a bicycle 118. The connector 116 has bracket 402 that connects to a bicycle seat 118. The bracket 402 is arranged to lock onto both of a pair of standardised seat-post-connection bars of a bicycle saddle undercarriage 404, such that the connection between the bracket 402 and the seat-post-connection bars governs the orientation of the device 101 relative to the saddle. The battery 120 is used to the various components power the device 101, such as the projector 102, the camera 104 and the computer processor 108 as well as the other components of the device 101 mentioned above. The battery 120 may be USB- rechargeable battery or any other suitable battery or power source as is known in the art. Figure 2 is a flow diagram of a method 200 for road safety. The method 200 comprises six steps. The first step 202 of projecting a pattern of light on a section of a road. The second step 204 of capturing one or more images of the pattern of light projected on the section of road and an overtaking vehicle 304, the camera 104 and projector 102 having a predetermined difference in viewing direction of the section of road. The third step 206 of analysing distortions in the pattern of light in the captured one or more images. The fourth step 208 of mapping a surface of the section of the road based on the distortions of the pattern of light. The fifth step 210 of classifying one or more features of the overtaking vehicle 304. The sixth step 212 of determining a relative location between the device 101 and the overtaking vehicle 304 based on the map, the classified feature of the overtaking vehicle 304 and the predetermined difference in viewing direction. Figures 3A-C show perspective illustrations of an in-use scenario of a device 101. Figure 3A illustrates a bicycle 118 and an overtaking vehicle 304 traveling along a road surface 306. In this illustration, the overtaking vehicle 304 is approaching the bicycle 118. Both the overtaking vehicle 304 and the bicycle 118 are traveling in the same direction (as shown by the arrow 310 in Figure 3C). Figure 3B shows the bicycle 118 with the device 101 attached. It illustrates the projected pattern 308 that may be expected when the projector 102 is in operation as described above. In this example, the projected pattern 308 is parallel lines that have been deformed by the uneven surface of the road surface 306. Figure 3C illustrates the overtaking vehicle 304 approaching the bicycle 118 travelling along the same direction. The direction of travel is indicated by the arrow 310. The nearside of the overtaking vehicle 304 interacts with the projected pattern 308. Once the nearside tyre of the overtaking vehicle 304 is close enough the device 101 to interact the projected pattern 308 and be captured by the camera 104, the device 101 is able to determine the relative location, relative speed and / or the absolute speed of the overtaking vehicle 304 as described above. Figure 4 is an illustration of the device 101. Figure 4 shows the bicyclist 302 with a bicycle 118. The device 101 is attached to the bicycle 118. In this example, the device 101 is attached to the bicycle saddle undercarriage 404 with a bracket 402. Also illustrated in figure 4 is the projector 102 and the camera 104 of the device 101. The camera 104 and the projector 102 are a known and a fixed distance apart within the device 101. The camera 104 has a field of view as illustrated by the field of view arrows entering the camera 104. The projector 102 has an infrared LED light source 406 and a diffraction grating 408. The infrared LED 406 shines light through the diffraction grating 408 to produce the projected pattern 308 on the road surface 306. The projector 102 projects a projected pattern 308 in a direction that overlaps with the field of view of the camera as indicated by the arrows that are emitted from the projector 102. Figure 5 shows a nearside tyre 502 of an overtaking vehicle 304. The nearside tyre may be a front nearside tyre or a rear nearside tyre. Figure 5 illustrates how the device 101 can determine a notional contact position 504. The notional contact position 504 is a position that is proximate to the actual position where the tyre 502 contacts the road surface 306. As described above, the notional contact position 504 is used to determine the relative location between the device 101 and the tyre 502. The notional contact position 504 lies at, or approximately inboard of, the outboard extremity of the tyres 502 outer wall 506. An arrow 508 illustrates that the camera 104 and projector 102 would not be able to determine the actual contact position where a tyre contacts the road surface 306 when the device 101 is operated in normal use, this is due to the geometry of a standard tyre. To overcome this problem, the notional contact position 504, which is a position that is approximates the actual contact position, is used when the device 101 determines the relative distance between the device 101 and the overtaking vehicle 304; the relative speed of the device 101 and the overtaking vehicle 304 and actual speed of the overtaking vehicle 304. It will be understood that the present invention has been described above purely by way of example, and modification of detail can be made within the scope of the invention.

Claims

1. A device for road safety, comprising:- a projector configured to project in use a pattern of light on a section of a road;- a camera configured to capture in use one or more images of the pattern of light projected on the section of road and an overtaking vehicle, the camera and projector having a predetermined difference in viewing direction of the section of road;- A classifier module comprising a model that is trained to classify one or more features of a vehicle using the one or more captured image of the overtaking vehicle;- a computer processor configured to receive the one or more images, and to analyse distortions of the pattern of light, and further configured to:map, based on the distortions of the pattern of light, a surface of the section of the road;classify, using the classifier module, one or more features of the overtaking vehicle;determine, based on the map, the classified feature of the overtaking vehicle and the predetermined difference in viewing direction, a relative location between the device and the overtaking vehicle.

2. A device according to claim 1, wherein the device for road safety is for use on a user’s road vehicle having an axis of travel, and in an in-use configuration, the relative location is determined between the device and the overtaking vehicle, is a lateral distance with respect to that axis of travel.

3. A device according to claim 1 or 2, wherein the projector is configured to project light in the infrared spectrum or near infrared spectrum.

4. A device according to any preceding claim, wherein the pattern of light is generated by directing a laser or LED through an optical element.

5. A device according to any preceding claim, wherein the pattern of light is a two-dimensional pattern.

6. A device according to claim 5, wherein the two-dimensional pattern is a line pattern.

7. A device according to any preceding claim, wherein the camera is orientated to capture a front, or a rear, nearside tyre of the overtaking vehicle.

8. A device according to any preceding claim, wherein the camera is orientated to capture a vehicle number plate of the overtaking vehicle.

9. A device according to any preceding claim, wherein the model is a trained to classify a front, or a rear, nearside tyre of a vehicle.10.A device according to any preceding claim, further comprising an automated number plate recognition system.

11. A device according to any preceding claim, wherein the relative location between the device and the one or more features of the overtaking vehicle is determined by:- determining, using the captured one or more images of the front or rear nearside tyre and the map of the surface of the road, a notional contact position that lies at, or proximately inboard of, the outboard extremity of that tyre’s outer wall; and- determining the relative location between the notional contact position and the device.

12. A device according to claim 11, wherein the notional contact position is a point on the periphery of the tyre that is nearest the mapped road surface from the viewpoint of the camera.13.A device according to any preceding claim, wherein the camera is configured to capture in use at least two images of the overtaking vehicle and the computer processer is configured to calculate a relative velocity between the device and the overtaking vehicle; wherein the relative velocity is calculated by determining the relative locations of the overtaking vehicle and the device over a period of time.

14. A device according to claim 13, wherein the computer processor is configured to receive the absolute travelling velocity of the device and to calculate, using the absolute velocity of the device and the calculated relative velocity, the absolute velocity of the overtaking vehicle.

15. A device according to any preceding claim, wherein the device further comprises a light source that is either distinct from the projector or operates in a different wavelength band, configured to substantially evenly illuminate at least one part the section of the road.

16. A device according to claim 15 when dependent on claims 7 and 9, wherein the light source is configured to illuminate at least part of the section of the road corresponding in use to a direction where a vehicle tyre is to be measured, with a greater radiant light intensity compared to the radiant illumination another part of the section of road corresponding in use to a direction where a numberplate of the overtaking vehicle is to be captured.

17. A device according to any preceding claim, wherein the device further comprises at least one additional sensor for independently measuring the relative location, or the relative location and vehicle velocity, wherein the at least one of such additional sensor comprises an infra-red, ultrasonic, or radar distance sensor.

18. A device according to any preceding claim, the device further comprises a red light oriented to the rear in an in-use position on a bicycle.

19. A device according to claim 18, wherein the red light is in the form of a symbol or an image with time-varying shape or motion, such as to present a sign to approaching motorists.20.A device according to any preceding claim, further comprising a connector for attachment to a bicycle.21 .A device according to claim 20, wherein the connector is configured to attach to the seat post or seat of the bicycle.

22. A method for road safety, comprising the steps of:- projecting a pattern of light on a section of a road;- capturing one or more images of the pattern of light projected on the section of road and an overtaking vehicle, the camera and projector having a predetermined difference in viewing direction of the section of road;- analysing distortions in the pattern of light in the captured one or more images;- mapping a surface of the section of the road based on the distortions of the pattern of light;- classifying one or more features of the overtaking vehicle;- determining a relative location between the device and the overtaking vehicle based on the map, the classified feature of the overtaking vehicle and the predetermined difference in viewing direction.

23. A method according to claim 22, wherein the camera is orientated to capture a front or a rear nearside tyre of the overtaking vehicle.

24. A method according to claim 23, wherein determining the relative location between the device and the overtaking vehicle further comprises:- determining, using the captured one or more images of the front or rear nearside tyre and the map of the surface of the road, a notional contact position that lies at, or proximately inboard of, the outboard extremity of that tyre’s outer wall; and- determining the relative location between the notional contact position and the device.

25. A method according to claims 22 to 24, comprising capturing in use at least two images of the overtaking vehicle and calculating a relative velocity between the device and the overtaking vehicle; wherein the relative velocity is calculated by determining the relative locations of the overtaking vehicle and the device over a period of time.

Images