SDW - Safe Distance Warning - Revision history

Abilotti at 17:44, 8 July 2020

2020-07-08T17:44:00Z

Abilotti at 19:34, 6 July 2020

2020-07-06T19:34:27Z

Abilotti at 18:00, 6 July 2020

2020-07-06T18:00:06Z

Abilotti at 16:56, 6 July 2020

2020-07-06T16:56:32Z

Abilotti at 16:42, 6 July 2020

2020-07-06T16:42:23Z

Abilotti: Created page with "Safe Distance Warning (SDW)feature is processing images directly on-device, without any cloud connectivity or additional add-ons. SDW feature is part of outside camera featur..."

2020-07-06T16:19:06Z

Created page with "Safe Distance Warning (SDW)feature is processing images directly on-device, without any cloud connectivity or additional add-ons. SDW feature is part of outside camera featur..."

New page

Safe Distance Warning (SDW)feature is processing images directly on-device, without any cloud connectivity or additional add-ons.

SDW feature is part of outside camera features and is running at 4 FPS and at 2 FPS if inside camera AI features are turned on. The algorithm triggers an event called “SDW” when a specific vehicle speed/distance to the car in front report is reached. The event is uploaded on Cloud on cellular data by default.

'''SDW architecture of this feature consists of:'''
#'''A speed/accuracy optimized state of the art Convolution Neural Network (CNN) which is:'''
#:*detecting all the vehicles in the field of view;
#:*assessing vehicle types;
#:*assessing distance through the vehicle.
#A deep neural network which is processing the vehicles from the 1st CNN network and based on their characteristics, type, distance and an offset calibration parameter (distance from the camera to the center of the windshield) is assessing if a vehicle is in front of our car or in a side position (other lane).

3. Knowing your car current speed (from GPS) and distance to the closest car in front, we can determine if you are keeping a safe distance to the car in front or not.

Very important!!!
In order to SDW to function properly, the position of the camera needs to be calibrated. Calibration is made through Iris Fleet Manager where the client or the support staff will have to write if the camera is in the right side or the left side of the windshield and also the distance from the middle vertical axis of the AI camera to the center of the mirror. The distance can be in inch or cm depending on the user interface settings.

Settings
The default settings for vehicle speed/distance to the car in front report are in the table below, but every client can customize the thresholds depending on their needs by informing support which will make a request to engineering:

Distance class Category Distance in m and feet Speed threshold on your car for making an SDW event
Extremely close frontal vehicle 0-3 m (0 feet -9 feet and 10) 40 km / h (24.8 mph)
Close frontal vehicle 3-6 m (9 feet and 10 - 19 feet and 8) 70 km / h (43.5 mph)
Midrange frontal vehicle 6-10 m (19 feet and 8- 32 feet and 9) 85 km / h (52.8 mph)
Far-mid range frontal vehicle 10-15 m (32 feet and 9 - 49 feet and 2) 120 km / h (74.5 mph)
Far frontal vehicle Over 15 m (over 49 feet and 2) 140 km / h (87 mph)

Safe Distance warning accuracy is 96% during the day when the camera in mounted on the center of the car. On more than 10 cm offset from the center of the car the accuracy drops to 92%.
During the night, the accuracy is 88% when the camera in mounted on the center of the car. On more than 10 cm offset from the center of the car the accuracy drops to 85%.
If false detections are present in a large quantity on a client, please inform engineering for adjusting the confidence thresholds.

Technical recommendation:
For a very good precision of this feature mount the camera as closest to the center as possible.

Other information:
As a safety method, a positive detection will need to be validated on 2 consecutive frames in order to give a trigger. So, the vehicle will need to be between 0.5 - 1 seconds in a dangerous position for an event to be triggered.

@@ Line 9: / Line 9: @@
 #:*assessing vehicle types;
 #:*assessing distance through the vehicle.
-#A deep neural network which is processing the vehicles from the 1st CNN network and based on their characteristics
+#'''A deep neural network which is processing the vehicles from the 1st CNN network and based on their characteristics'''
 #*type
 #*distance
 #*and an offset calibration parameter (distance from the camera to the center of the windshield) is assessing if a vehicle is in front of our car or in a side position (other lane).
-#Knowing your car current speed (from GPS) and distance to the closest car in front:
+#'''Knowing your car current speed (from GPS) and distance to the closest car in front:'''
 #*we can determine if you are keeping a safe distance to the car in front or not.

← Older revision		Revision as of 19:34, 6 July 2020
Line 65:		Line 65:
	'''Technical recommendation:'''		'''Technical recommendation:'''

−	Mounting the camera as close to the center as possible will provide the highest ~~precision~~ of this feature	+	Mounting the camera as close to the center as possible will provide the highest Accuracy of this feature

@@ Line 25: / Line 25: @@
 *also the distance from the middle vertical axis of the AI camera to the center of the mirror.
-[[File:SDW mount config.png]]
+[[File:SDW mount config.png]] '''Note:'''
 "Distance from center" is displayed as inch(s) or cm(s) depending on the user interface settings.
@@ Line 35: / Line 32: @@
 Default settings for vehicle speed/distance to the car in front report are in the table below, but every client can customize the thresholds depending on their needs by informing support which will make a request to engineering:
-Safe Distance warning accuracy is 96% during the day when the camera in mounted on the center of the car. On more than 10 cm offset from the center of the car the accuracy drops to 92%.
+*Safe Distance warning accuracy is 96% during the day when the camera in mounted on the center of the car.
-During the night, the accuracy is 88% when the camera in mounted on the center of the car. On more than 10 cm offset from the center of the car the accuracy drops to 85%.
+**On more than 10 cm offset from the center of the car the accuracy drops to 92%.
 If false detections are present in a large quantity on a client, please inform engineering for adjusting the confidence thresholds.
-Technical recommendation:
+'''Technical recommendation:'''
-For a very good precision of this feature mount the camera as closest to the center as possible.
-Other information:
+<br/><br/>

@@ Line 24: / Line 24: @@
 *Calibration is made through Iris Fleet Manager where the client or the support staff will have to select if the camera is in the right side or the left side of the windshield
 *also the distance from the middle vertical axis of the AI camera to the center of the mirror.
 '''Note:'''
-The distance is shown as inch(s) or cm(s) depending on the user interface settings.
+"Distance from center" is displayed as inch(s) or cm(s) depending on the user interface settings.
-[[File:SDW mount config.png]]
+Default settings for vehicle speed/distance to the car in front report are in the table below, but every client can customize the thresholds depending on their needs by informing support which will make a request to engineering:
 Distance class Category	Distance in m and feet	Speed threshold on your car for making an SDW event

@@ Line 1: / Line 1: @@
-Safe Distance Warning (SDW)feature is processing images directly on-device, without any cloud connectivity or additional add-ons.
+'''Safe Distance Warning (SDW)''' feature is processing images directly on-device, without any cloud connectivity or additional add-ons.
 SDW feature is part of outside camera features and is running at 4 FPS and at 2 FPS if inside camera AI features are turned on. The algorithm triggers an event called “SDW” when a specific vehicle speed/distance to the car in front report is reached. The event is uploaded on Cloud on cellular data by default.
 '''SDW architecture of this feature consists of:'''
@@ Line 8: / Line 9: @@
 #:*assessing vehicle types;
 #:*assessing distance through the vehicle.
-#A deep neural network which is processing the vehicles from the 1st CNN network and based on their characteristics, type, distance and an offset calibration parameter (distance from the camera to the center of the windshield) is assessing if a vehicle is in front of our car or in a side position (other lane).
+#A deep neural network which is processing the vehicles from the 1st CNN network and based on their characteristics
-Very important!!!
+In order to SDW to function properly, the position of the camera needs to be calibrated