Automotive

Object detection software to enhance driver awareness, improve safety

View 3 Images
The new AOD object-detection software system from Fujitsu
The AOD process flow
AOD Display
The new AOD object-detection software system from Fujitsu
View gallery - 3 images

When it comes to driver awareness, we all know how hard it can be to keep an eye on every pedestrian and moving vehicle in our vicinity, particularly when driving in a busy city area. Couple this with the upcoming Kids and Transportation Safety Act in the US that dictates that all cars built from May 2018 onwards must be fitted with rear view cameras as standard, and it becomes doubly important that on-board vehicle cameras are used to their best advantage. To help in this regard, Fujitsu Semiconductor Limited is set to introduce software that assists in detecting and identifying cars, people, and other moving objects and alerts the driver of their position and direction of travel.

Using what Fujitsu calls its "Approaching Object Detection Library" in conjunction with a proprietary System-on-Chip (SoC) graphics device, the system uses the images captured by additional vehicle cameras to automatically detect nearby moving objects and compare them to stored representations to identify and categorize them.

This is achieved by using elements built into the Approaching Object Detection Library, where vehicle camera images are analyzed in conjunction with a detection algorithm that identifies approaching objects. These are then run with detection-error reduction processing to eliminate false positives. The resulting image is overlaid on real-time images collected through the vehicle's on-board cameras, and displayed on a dashboard monitor, providing vital information for the driver about the objects moving around them.

Though there are other systems available that aim to improve a driver's all-round vision, they still require constant monitoring and, given the increasing demands on drivers with all of the gadgets that vie for attention within the car, the AOD system may well do even more good simply by being unobtrusive.

The AOD library will be available in an evaluation version from this month, and will be shipped with an evaluation board that includes Fujitsu's graphics SoC. Release for commercial sale is slated for September this year.

Fujitsu will also be demonstrating this and other solutions at the 2014 Automotive Engineering Exposition, May 21–23 in Yokohama, Japan.

Source: Fujitsu

View gallery - 3 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
5 comments
Slowburn
But officer my car did not show me any pedestrians.
Mel Tisdale
At least there are mark 1 human eyeballs connected to a human brain that has ultimate decision making responsibility. Google, however, seem to be quite happy that those human components could be asleep or distracted by a text message or Sudoku puzzle, etc. It is obviously going to take a WTF moment to bring them to their senses. I just hope that that moment happens before a body bag is required. Perhaps Fujitsu and Google could share their technologies.
Sheldon Cooper
So the driver has to look down at a dashboard monitor to 'see' moving objects? How about just looking through the windscreen?
Anything which encourages the abdication of responsibility for being alert and observant is bad.
Stephen N Russell
Lisc for other auto makers to produce & use. Awesome Install for 2016 models forward.
Daishi
@Sheldon Building a working system able to identify dangers is the hard part. Doing something with it is easier. If you are looking away and the car in front of you stops suddenly, do you need to check an LCD to know?
No, the car could warn you and begin to decelerate. Even when people are paying attention they can be bad at identifying some hazards like motorcycles. Look at 2:30 of this video as an example www.youtube.com/watch?v=eqQBubilSXU#t=150
Things like pillars and passengers can also obstruct your view so having an extra set of "eyes" can be a good thing.
Cheaper 3D imaging like Kinect and Project Tango combined with mobile GPU computing (like Nvidia CUDA) will allow the hardware for this to be created cheaply.
The Air Force was using PS3 as a supercomputing platform to analyze satellite images because they were 200-400 GFLOPS each and now there are faster mobile GPUs. The R9 295X2 GPU is 11.5 TFLOPS. Pretty soon cars will have more processing power than supercomputers from a few years ago.