Technology

On and off road: unbeatable image processing ability

iStock-olaser - ok -

Mechanically stronger so it can operate in harsh environments, Blaxtair’s pedestrian detection solution is also more effective at processing and analyzing images than the technology used currently in the automotive industry. It is the result of the highest levels of scientific and technical collaboration with the CEA (The French Atomic Energy and Alternative Energy Commission) as their head of Machine Vision Systems, Laurent Lucat, from the List institute, explains[1].

How is the pedestrian detection technology used in the automotive industry different from Blaxtair’s system?

It is mainly to do with the technological implications associated with its context of use. Most systems used in the automotive industry need to detect a pedestrian, without needing to know their precise location, standing in or along the side of a road, that is, on a relatively flat and consistent surface. Processing the image is then simpler since we know where to look for the pedestrian in the image – by taking in an average angle and average size. In this case, a monocular system connected to the CAN bus that contains the driving information and, primarily, the speed (the car must be moving) is enough.

Why does Blaxtair need “two eyes”?

Again, it is a question of context. Worksite machinery does not necessarily work on flat surfaces. So to detect a pedestrian, it has to be able to analyze the ground. It also has to identify much more precise detection zones because the system does not need to sound an alert every time it detects a pedestrian, but only when that pedestrian is in a danger zone. That is why we use 3D stereoscopic vision. Also, unlike in the case of cars where trajectories are relatively constant, in the context of worksite machinery, we cannot rely on image continuity to consolidate the information over time: take forklift trucks, for example, their fast, pivoting turns mean that images change suddenly, and it has to be capable of interpreting them. It is therefore necessary to gain the maximum amount of information possible from each image.

How is such a high level of image processing and analysis ability achieved?

The List institute has been developing algorithms for around ten years now, that are increasingly taking into account industrial contexts. It is the context learning that gives this technology such strong image analysis ability and what allows us to account for irregular terrain in real time, for example. Today, Blaxtair can detect 99% of pedestrians in a danger zone and we are continuing to improve this performance by constantly integrating new research results or new contextual elements encountered by Blaxtair whilst being used by its clients.

The List institute

List, a CEA Tech institute and head of technological research for the CEA, focuses its research on digital intelligent systems. In response to major economic and societal challenges, its R&D programs center on advanced manufacturing, embedded systems, ambient intelligence and ionizing radiation for healthcare applications. By developing state of the art technologies whose uses span the transport sector, security/defense, manufacturing, energy, health and ICT, the Carnot CEA LIST institute contributes to its partners’ industrial competitiveness through their innovation and technological transfer (www-list.cea.fr).

[1] List is a CEA Tech institute, heading up technological research at the CEA.

24 June 2016

Imprimer
Share