From 4 to 6 July, the Cloud LSVA consortia met at Dublin City University (DCU) to prepare the final integration of the Cloud LSVA results in several appealing demonstrations, which clearly validate the goals and achievements of the project over the last three years.
Agreements and conclusions were made both on a technological level and on the remaining activities for the last 6 months of the project. The first two days were dedicated to the core technical aspects of the project and the synergies between the different work packages and tasks. During these discussions several very interesting results of the project were demonstrated, in particular the real-time annotation of pedestrians.
The main goal for pedestrian annotation was set to be the ability to detect pedestrians and annotate objects in 3D by drawing the skeleton structure of the person and synchronizing multiple cameras, fish-eye images and stitching together panorama cameras to collect as many details as possible, both from the pedestrians and the surrounding environment. By doing so, it is be possible to collect more details, and therefore provide more accurate images.
The presentations demonstrated that there has been indeed significant progress in the field: at this stage of development, the detection and recording of attributes of a person, such as the gender, the age and much more, is possible thanks to the enhanced annotation by means of polylines of scenery elements.
This technology is also very important to predict certain movements, such as crossing the street or seeing when a pedestrian is walking near the vehicle.
In terms of collecting images from the surrounding environment, the ultimate goal would be to crowdsource traffic sign locations, using pictures from dash cameras and being able to send images observed from the car to the cloud in real time and to be subsequently shared among other autonomous vehicles.
What was very important after the meeting was a cleared definition of the results to present during the 2018 ITS World Congress, which will take place in Copenhagen from 17 to 25 September, and during the final event. Near the end of November, the project will organise a final event which will include a full demonstration of the achieved results and a hand -on possibility to test the developed components. More information will follow as soon as a concrete date and location is set.
The last day of the consortium meeting was dedicated to a small workshop focussing on the business and exploitation aspects of the project. All experts agreed on the importance of large video scale annotation as a research which can be applied on a cross-sectoral scale. In fact, video annotation can benefit various domains, from automation to robotics, indoor outdoor environments, security, insurance, law enforcement agriculture (for example by using drones) the medical sector and anything that generates visual data.
With many development possibilities ahead, don’t miss the next updates on the Cloud LSVA project. The last few months promise to be very exiting with all jigsaw pieces coming together in the final project deliverables. It is clear that the Cloud LSVA results provide a huge step forward in the recognition of the near vehicle scenery, a technology which is indispensable for the new type of semi- and full automated vehicles which will hit the road in a near and somewhat further future.