In Vehicle Implementation Experience: Result Analysis
We faced several challenges when trying to integrate our lab-developed and tested solution to a real truck. Mechanical limitations (PC location, attachment possibilities, ventilation conditions) and electrical installation concerns led us to make a limited and controlled test (15 minutes’ time driving around the neighborhood). The truck used for the test is used for small, short range freight services. In order not to create hazard due to the installation of the device permanently and not to impact the business, we decided to go with a removable installation, for which we took several shortcuts that in a product-long term installation would not have been possible.
What follows is a summary of the setup and conditions under which we ran the test.
The in-vehicle implementation experience allowed us to validate several aspects related to the System Integration, that can be described as follow:
There was limited space available to install the computer (under the passenger seat only) with few fixation possibilities and scarce air flow. We decided to place the computer on the floor mat, monitored by one of the engineers during the test.
- 12V and ignition connections were made to power Nexcom ATC8010-7A computer used during the test. We utilized a start-up script for the automatic setup of the Driver Assistance system initialization environment as well as the system’s automatic suspension whenever the truck’s ignition went off.
- Internet Connection– In order to ensure connectivity throughout the duration of the test, we decided to connect to the internet by using a Wi-Fi USB Dongle connected to a Cellular acting as a Wifi Hotspot.
-
GPS Connection – Nexcom’s ATC8010-7A native GPS was used as the provider of the truck’s positioning service.
-
Camera Connection – Two USB cameras were used to detect driver’s fatigue and actions:
- Driver Recognition + Drowsiness/Distraction Detection: The USB camera was placed on the dashboard, behind the steering wheel (80cm – 100cm approximate away from Driver’s head).
- Driver Action Recognition: The USB camera was placed on the dashboard in front of the passenger seat, facing the driver’s position, allowing to have a wide capture of the driver’s complete upper body position.
- CAN Connection: The truck used for the test lacked a standard OBD-2 connector, thus the CAN bus information was replaced by a simulation.
After the device was successfully powered inside the truck, we checked the position of the cameras and started up the OpenVINO environment. Once the models were able to work with the cameras, we were able to check internet connection and the dashboard in the cloud. We made an end to end test. The driver entered the truck and started the engine. Then he drove for about 15 minutes in the neighborhood. During the time of the test, we could check the dashboard real time indicating the truck’s location, and the drowsiness and distraction indicators.
In-Vehicle Implementation video footage
When the ride ended, we checked that the system went down gracefully after ignition off.
- Initialization script ran as expected, configuring all the required preconditions for running Driver Assistance system:
- OpenVINO environment variables
- 4G/3G connection.
- Driver Assistance initialization
- Driver Action Detection configuration
- AWS Dashboard display
-
OpenVINO detection models worked as expected.
-
Detection process transition from Driver behavior to Driver Action worked as expected.
-
AWS cloud integration displayed events in real time:
- Drowsiness/Distraction Indicators
- Simulated Vehicle information
- Driver’s behavior events historical charts
- Driver’s actions description.
The in-vehicle implementation end to end was successful. The use cases exercised worked with a high degree of confidence:
- Driver recognition: the driver was recognized immediately after the driver’s face was framed by the front camera
- Drowsiness: the driver simulated yaws and blinks as well as head bumps which were identified. The drowsiness indicator behaved properly
- Driver distraction: whenever the driver took his eyes off the road, the system detected this situation. The driver’s behavior model could identify most of the distracting activities performed by the driver (mainly eyes off the road and cell-phone usage).
The use of OpenVINO for driver management allows the fleet manager to have real time information about the driver and the truck, and use this information for route analysis and/or driver’s assessment.
- Integrating the Nexcom’s ATC8010-7A device to the truck’s power line did not affect the truck’s electrical performance during ignition on/off cycles.
- The device did not overheat during the time spent for the in-vehicle implementation.
- When considering a permanent installation, forced air circulation on the device is recommended.
- Using a Wi-Fi hotspot for 4G/3G connectivity provided us better connection stability than USB 4G/3G modem for data streaming to AWS cloud dashboards ensuring the real time vehicle monitoring.
- Data local storage due to lack of connectivity was not in scope for this trial.
- The GPS connection was successfully integrated to the vehicle monitoring dashboard (AWS cloud) allowing to highlight the geographic position of the truck in real time.
- This feature should be adjusted to highlight events in the map display only when the behavior values are “significant” (over 30 = green scale at the Vu Meter).
