GB2527330A - Railway vehicle position and specific track location, provided by track and point detection combined with optical flow, using a near-infra-red video camera - Google Patents

Abstract

A railway vehicle position determination device enabling a railway vehicle to accurately measure its position on the railway and, additionally, to report the precise track the vehicle is currently running on. The invention uses a near-infra-red (nIR) digital video camera to detect the current rail track the vehicle is running on and also to detect when a rail point, such as a track junction, is passed and on which track the rail vehicle occupies following the rail point. The speed of the rail vehicle is measured to give a measurement of how far the train has travelled since it last passed a rail point. This device reports, with respect to a railway system map, where the vehicle is on the track, both longitudinally and transversely, so as to provide precise track identification.

Description

Tide: Railway vehicle position and specific track location, provided by track and point detection combined with optical flow, using a near-infra-red video camera

Background:

The rail industry rely upon infrastructure, such as track circuits and axle counters, to measure where each train is. These systems do not always provide continuous updates and are prone to failure or theft. Consequently when a piece of infrastructure fails it can bring the whole rail network to a halt.

Rail workers are deployed to the track-side to monitor the rail traffic and communicate with the signal operator. The track-side workers manually signal to train drivers using flags in order to prevent accidents from occurring. Therefore a positioning system which can provide the rail operator and signal operators with the precise position of a train, when such failures occur, can help prevent or completely avoid disruption to the rail network. Global Positioning System (GPS) is one method of tracking rail vehicle position, however, this does not reliably or definitively provide an accurate identification of which actual track each rail vehicle is currently on. As such train drivers are contacted and rail workers deployed to the track side to report the position of each train to the signal and rail operators. This invention provides an accurate way to determine where a train is and which track it is on without the need for deploying extra human resources or querying the train driver.

This invention provides a mechanism for a railway vehicle to accurately measure its position on the railway, and more importantly report the actual track the vehicle is currently running on. The invention requires a near-infra-red (nIR) digital video camera to detect the current rail track the vehicle is running on, detect when a set of rail points is passed, and track the vehicles movement through those points. The distance travelled and speed of the vehicle is measured to give a measurement of how far the train has travelled since it last past a set of points. This invention reports, with respect to a prior-map, precisely where the vehicle is on the track, both longitudinally and transversely giving the precise track identification.

This information is collected by the vehicle and can be reported back to headquarters to provide another system by which the railway operators are able to know where the trains are when GPS and railway infrastructure fail, or where no other method of positioning is possible. For example in a tunnel.

Statement of invention:

A railway vehicle is fitted with either a forward and rear facing near-infra-red (fIR) video camera.

The rails are highly visible within this spectrum in all weather conditions and are illuminated using non-intrusive invisible infra-red light at night or when it is dark. The images of these cameras are then processed and a rail detector used to detect all currently visible rails. The system then detects when the rail vehicle has passed a set of points due to rails either appearing to merge or diverge.

The current rail is tracked and therefore the route through the points is also tracked and whether the current rails merge with another set. The video camera feed is also processed to output Optical flow which enables the system to determine how far it has travelled between consecutive video frames and consequently also provide a distance travelled and speed measurement. The system uses the distance travelled combined with the tracking of tile vehicle through rail-points to determine the vehicles position within a prior-map of the railway.

Advantages: * Accurate and repeatable position estimate relative to railway map.

* Provides position, speed and track identification output in all weather conditions providing camera lens is not obscured.

* Using infra-red illumination and near-infra-red camera results in rails being visible in all weather conditions in which rail vehicles are legally allowed to operate.

* Other systems rely on providing position information by tracking obj ects and infrastructure which are not the actual infrastructure the vehicle is running on, i.e. the rails, and as such they can be subj ect change and could result in false readings. This system requires only the rails themselves as it is these which are tracked.

* Since knowing which specific track a rail vehicle is travelling on is very important, this system definitively reports the movements between rails through points by tracking the points themselves. The tracking of no other object i.e. GPS, signposts, QR codes, can definitively and instinctively provide a reliable result for which set of rails a railway vehicle is currently travelling upon. The rails are the most reliable object to track and not subject to sporadic change. For example, sign posts with signs on can be blown over or be obscured by snow, however, the rails themselves are still visible in these conditions.

* Does not require prior map and will output distance travelled since last rail points, and also the list of point changes recorded. Therefore another system can use this data externally to position the vehicle providing it has a map.

* Requires no additional railway infrastructure to be installed.

* Velocity and distance travelled measured without requiring prior map.

* Provide pseudo GPS (Global Positioning System) location estimate when combined with prior map logical network map.

* Can work on any rail guided system such as those in factory automation, quarrying, mining and recreational toys.

* System can work with virtual rails such as painted white lines and therefore suitable for factory autom ation.

* System can also output speed, distance travelled and direction of movement when not using rail tracking feature.

Introduction to Drawings:

Figure 1 provides step-by-step examples of how to detect rails in images.

Figure 2 provides step-by-step examples of how optical flow works.

Figure 3 provides a step-by-step guide of how to combine the information from point detection, optical flow and a prior map to report position and track identification. This information can be correlated to GPS positions of the map to provide a pseudo-gps position estimate.

Figure 4 Explains how to capture rail track images with an near-infra-red camera and invisibly illuminate the rail track using infra-red light. This also shows the measurements required for manually converting from optical flow to distance travelled or speed. The angle [p1 is the optimum angle the camera pitches down to eliminate the horizon. Angles [x][y] refer to field of view of the camera. These angles along with the height are used to determine a manual conversion factor from optical flow to speed in meters per second.

Detailed Description:

A near-infra-red camera is fitted to the front or rear of a rail vehicle with an infra-red light also pointing in the same direction as the camera (See Figure 4). For best results the camera should be angled slightly downward so that only the rails are visible in the camera image and not lots of sky or horizon. The images are then processed by a computer.

The processing steps are as follows: 1. Convert image to luminosity image, i.e. not colour.

2. Rectify and de-warp the image to take account of angle of mounting relative to plane of the rails and to remove any fish-eye effects caused by the lens. This ensures that all straight lines in image appear straight. Figure 4 Shows required measurements and camera mounting position.

3. Run an edge filter over the image to highlight pixels with high intensity gradient and produce the gradient image. This should highlight edges such as rails.

4. Filter out long edges. See Figure 1.

5. Identify likely candidates for rail pairs. Can use string prior information about distance between each rail in a pair and maximum curve rate of a rail to cull false detections.

6. Identify current rail pair being travelled along by picking closest pair to camera origin.

7. Follow each rail detection from one image to the next. Giving each rail detection a temporary identification number.

8. When rails spilt (diverge), since this occurs at sets of points, track the current rail the train seems to be travelling on by picking the closest rail detection to the known camera position.

The direction the new rail appears to move away from the current rail is the opposite direction to which the train went at the set of points. For example if a new rail appears to move laterally to the right in the image the rail the train is on must be the left rail, meaning the train went left at the set of points.

9. The joining of the current rail to another is measured by detecting when two rails merge into one. The track point is detected when the rails have merged into the same position and only the new rail track identifier remains (that of the new rail being merged onto).

10. The track point information is reported as the dlirection the train went at each point when the rails split or the direction the train merged from onto the new rail when rails merge. Or if rails merged on current track the direction those rails joined from. e.g. a track point traversal history could look like: right, right, merge left, merge right, merge right. See Figure 3.

11. Optical flow is calculated by tracking the mutual information of consecutive image frames of the rectified, perspective corrected images. For example, how far a group of pixels have moved from one image to the next. See Figure 2.

12. Optical flow in pixels is then converted to metric distance travefled (meters) and speed (meters per second). The conversion factor can be automatically calibrated using a GPS if available or can be calculated using mounting angle of the camera relative to the ground plane and the height of the camera relative to the ground plane. The camera intrinsic calibration is required for the projection transform to be calculated.

13. The distance traveiled data is reset at each point detection to indicate distance travelled since last known set of points.

14. The point detection and directions are tracked with respect to the prior map. The map contains known distances between each set of points. This allows for the distance travelled, collected by optical flow, to be correlated to a position within the map.

15. If the prior map also contains measurements data of the tracks in its own coordinate system (x,y,z), for example Global Positioning System (GPS), then the distance travelled between a set of know points in the prior map can be converted to a pseudo GPS position. This allows the data to be easily compared with surveys containing other data, such as aerial photography or laser scan.

Claims (9)

1. Claims: 1. Rail track detection in camera images.
2. 2. Rail point detection in camera images using output from (1).
3. 3. Rail point direction tracking in camera images using out put from (1).
4. 4. Detecting which way vehicle went at track points using near-infra-red camera images.
5. 5. Reporting direction vehicle went at track-points detected in camera images.
6. 6. Rail track and rail point detection using non-visible spectrum near-infra-red camera images with infra-red light source to prevent driver dazzle and distraction and to detect rails in afl legally operable weather conditions for trains.
7. 7. Measures distance travelled since last rail point using optical flow of video in near-infra-red spectrum combined with (4).
8. 8. Position and specific rail track reported with respect to prior logical network map using output from (7).
9. 9. Position can be reported in relative track coordinate space or in the coordinate system of prior map.Amendment to the claims have been filed as follows Claims: 1. Rail track detection in camera images.2. Rail point detection in camera images using output from (1).3. Rail point direction tracking in camera images using out put from (1).4. Detecting which way vehicle went at track points using near-infra-red camera images.5. Reporting direction vehicle went at track-points detected in camera images.6. Rail track and rail point detection using non-visible spectrum near-infra-red camera images with infra-red light source to prevent driver dazzle and distraction and to detect rails in all legally operable weather conditions for trains.7. Measures distance travelled since last rail point using optical flow of video in near4nfra-red spectrum combined with (4).8. Position and specific rail track reported with respect to Prior-map using output from (7).9. Position can be reported in relative track coordinate space or in the coordinate system of Prior-map. IC) IC) a) r