Point cloud rendering of a car

[ solutions ]

RailView

Long-range obstacle detection for trains up to 1,000 meters ahead.

Point cloud rendering of a car

[ solutions ]

RailView

Long-range obstacle detection for trains up to 1,000 meters ahead.

NODAR RailView is a collision warning system designed for the rail industry to prevent collisions with objects up to 1000 meters away. The system utilizes customized software built on our HDK (NDR-HDK-2.0-100-10-A). With its ultrawide-baseline stereo vision and 3D sensing capabilities, RailView can detect a wide range of objects—beyond just people and cars—such as random trash, rocks of various shapes, and debris.

Thanks to NODAR's advanced autocalibration algorithms, the ultrawide-baseline stereo camera always stays calibrated, even with platform shock and vibration. By measuring the three-dimensional structure of the area ahead of the train, the system can identify any physical object above the track surface, even if it encounters the object for the first time. For added reliability, we incorporate state-of-the-art AI detection methods.

System Architecture


Core System. HDK (NDR-HDK-2.0-100-10-A)

  1. Computational module: industrial edge AI processing unit, Nvidia Orin AGX, fanless. 100-240 VAC, 50/60 Hz, 4A

  2. Sensor module: Ultrawide-baseline stereo vision camera, 1-m baseline, 10-deg HFOV

  3. (Add-on) Sensor module: Monocular thermal camera with state-of-the-art AI detector

Interfaces

(Add-on) Audio Alert System. USB-A to 3.5-mm audio adapter
(Add-on) Display module. HDMI interface

Navigation

(Add-on) Navigation module. USB GNSS module

Connectivity

(Add-on) 5G/LTE modem. 5G to Gigabit Ethernet Converter

Integration Hardware

  1. Cables and interfacing equipment

  2. (Custom) Hardware for mounting and cabling. We recommend attaching visible cameras to inside of windshield behind the wipers.

  3. (Add-on) DC-DC converter. Specify input supply voltage: 110V for electric locomotive and 72V for diesel locomotive. Output supply voltage: 24V, 9.2A, 221W max.

Specifications


Functionality

RailView pre-warns the locomotive driver about the presence of an obstacle, its size and, when possible, its type (aka class). The classes supported are

  • Human.

  • Animal. Specifically: bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, and giraffe

  • Road vehicle. Specifically: car, motorcycle, bus, truck, and bicycle

  • Train.

  • (add-on) Boulders.

RailView generates an audio alert signal indicating the presence of objects when objects are within the warning zone. The warning zone is specified with respect to the sensor location. The size of the warning zone is precomputed by the user based on the speed, stopping distance, and assumed reaction time of the locomotive pilot.

RailView provides railway track visualization with NODAR Viewer to view the camera images, the depth maps, the point clouds, and store data to disk.

Edge level analytical capabilities. The RailView platform is modular and allows for future upgrades to the analytical capabilities. For example, track damage assessment, track quality, tree overhang monitoring, and customer counting can be added.

Alert to locomotive pilot. Real-time alerts, both visual and audable. Audio alert is disabled when the locomotive is at zero speed (0 km/h).

Capabilities

Curved tracks. Track curvature is limited to the horizontal field of view of the lens.
Night view. Thermal and visible cameras with headlights.
Heavy rain, fog, or storm. Visible camera range is approximately equal to the meterological visibility.

Compliance

Visible cameras:
• IP67 (IEC standard 60529, figure 14) dictates that the camera must have no ingress of dust (dust proof) and be water proof up to 1 meter (3 ft 3 in) in depth for 30 min. Not hermetically sealed.
• IEC / DIN EN 60068-2-27 Environmental testing – Part 2-27: Tests – Test Ea and guidance: Shock. IEC / DIN EN 60068-2-64 Part 2-64: Tests – Test Fh: Vibration, broadband random and guidance
• IEC / DIN EN 60068-2-6 Environmental testing – Part 2-6: Tests – Test Fc: Vibration (sinusoidal)
• IEC / DIN EN 61000-6-2 (Generic standards - Immunity standard for industrial environments)
• EMVA Standard 1288 - Standard for Characterization of Image Sensors and Cameras

Computer: CE / FCC Class A / UKCA

Signal Aspect Recognition

(Add-on) Detection of the nearest signal ahead in the direction of locomotive movement along the current track. Audio alert when red aspect, yellow aspect, double yellow aspect, or green aspect is identified.
(Add-on) Integration with the Fog Pass System for signal location data

Functionality

RailView pre-warns the locomotive driver about the presence of an obstacle, its size and, when possible, its type (aka class). The classes supported are

  • Human.

  • Animal. Specifically: bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, and giraffe

  • Road vehicle. Specifically: car, motorcycle, bus, truck, and bicycle

  • Train.

  • (add-on) Boulders.

RailView generates an audio alert signal indicating the presence of objects when objects are within the warning zone. The warning zone is specified with respect to the sensor location. The size of the warning zone is precomputed by the user based on the speed, stopping distance, and assumed reaction time of the locomotive pilot.

RailView provides railway track visualization with NODAR Viewer to view the camera images, the depth maps, the point clouds, and store data to disk.

Edge level analytical capabilities. The RailView platform is modular and allows for future upgrades to the analytical capabilities. For example, track damage assessment, track quality, tree overhang monitoring, and customer counting can be added.

Alert to locomotive pilot. Real-time alerts, both visual and audable. Audio alert is disabled when the locomotive is at zero speed (0 km/h).

Capabilities

Curved tracks. Track curvature is limited to the horizontal field of view of the lens.
Night view. Thermal and visible cameras with headlights.
Heavy rain, fog, or storm. Visible camera range is approximately equal to the meterological visibility.

Compliance

Visible cameras:
• IP67 (IEC standard 60529, figure 14) dictates that the camera must have no ingress of dust (dust proof) and be water proof up to 1 meter (3 ft 3 in) in depth for 30 min. Not hermetically sealed.
• IEC / DIN EN 60068-2-27 Environmental testing – Part 2-27: Tests – Test Ea and guidance: Shock. IEC / DIN EN 60068-2-64 Part 2-64: Tests – Test Fh: Vibration, broadband random and guidance
• IEC / DIN EN 60068-2-6 Environmental testing – Part 2-6: Tests – Test Fc: Vibration (sinusoidal)
• IEC / DIN EN 61000-6-2 (Generic standards - Immunity standard for industrial environments)
• EMVA Standard 1288 - Standard for Characterization of Image Sensors and Cameras

Computer: CE / FCC Class A / UKCA

Signal Aspect Recognition

(Add-on) Detection of the nearest signal ahead in the direction of locomotive movement along the current track. Audio alert when red aspect, yellow aspect, double yellow aspect, or green aspect is identified.
(Add-on) Integration with the Fog Pass System for signal location data

Detection Range

weather
condition

train speed

train
speed

Normal
Visibility

obstacle
detection range

Signal Aspect detection range

Clear

<160 km/h

Horizon limited

1000 m (VIS)

TBD

Mild Fog

<160 km/h

300 m

300 m (VIS)
500 m (LWIR)

TBD

Dense Fog

<95 km/h

100 m

100 m (VIS)
250 m (LWIR)

TBD

Extreme Fog

<20 km/h

5 m

5 m (VIS)
15 m (LWIR)

TBD

False positives: <3 per 100km

False negatives = 0 (over testing to date)

Detection Range vs. Object Size

The minimum detectable object size vs. range for the 5.4MP HDR VIS camera and 10-deg HFOV lens.


Range

Min Object Size

500 m

0.5 m

800 m

0.8 m

1000 m

1.0 m

Additional Specifications

Recording. The current rugged embedded system has a 500 GB internal SSD. An external USB SSD can be attached. Assuming a capacity of 4 TB, the raw uncompressed 5.4mp 16-bit images from the two visible cameras (108 MB/s), can be stored for 10.3 hours at 5 FPS.

  • (Add-on) Compression module to allow storage of 15 days of compressed video.

  • (Add-on) Built-in self-test on power up, and give warning to the driver if the system fails and is not functioning as expected.

  • Power up time is less than 2 minutes.

Deploy with NODAR