Dashcam Annotations

The last decades have seen steady improvements to road safety. Seat belts, ABS and Airbags have reduced the number of fatal road accidents. The next big improvements on road safety are systems that can automatically detect the potential for an accident and help the driver with automatic braking. This requires the on-board computer to understand the current scenery in front of the car by interpreting camera inputs. These systems are developed using a large amount of test images collected in datasets.

Currently available datasets are very much focused on middle/northern Europe in good conditions.

This means that solutions created with the existing data might not work reliably in other parts of the world. To improve on this situation we want to draw images from existing dashcam videos from all over the world. The resulting dataset will be made available for free to any interested researcher worldwide trying to improve road safety. If you access this page it normally means that we asked you for permission to use one or more of your videos in our dataset. Please consider supporting our project. All you have to do is reply with a simple “I Agree” and you can help to improve road safety globally.

Here are some answers to common questions:

• What are you actually asking me for?
  We are collecting individual frames from dashcam videos and want to include them in a dataset that will be publicly available. As an author of the original content (the videos): we ask you for permission to use individual frames under the CC-BY-NC 4.0 license ( https://creativecommons.org/licenses/by-nc/4.0/ ). We will collect a small subset of the frames and publish them as part of a dataset. The camera images of the dataset itself will also be released under CC-BY-NC. The release will contain a complete listing to attribute the original authors. Researchers can than use the images for non-commercial use. Publications based upon our dataset will have to reference the dataset license and our group as the collectors of the data. You will of course still retain all of the publication rights to your work and other rights connected to your authorship.
• Which actions/responses do you need from me?
  For for youtube messages: a simple reply with “I Agree”
  For Emails: Send a reply email which still includes our original request text and a simple “I Agree” as a reply from you is enough. Alternatively, feel free to reply with any questions you have about the release. This is a research project with very limited budget done by Phd. students. We have no budget to pay you for publication rights. Sorry! This is your chance to do a small step which can result in a huge impact on road safety! If you do not like to participate, please answer with a simple “No”.
• What data will be published within the dataset?
  We are publishing individual frames (i.e. images) from your video together with an annotation of the image. The annotations will be done by us and include categories like car; truck, road; sky; etc. To make the task easier for computer programs we will also publish  some of the previous and next frames for each selected frame. From longer sequences we might publish multiple timestamps. Each of these will be thoroughly spaced so that they do not show the same location. This means that your video cannot be reconstructed in any way. No audio track, video description text or discussion/forum texts will be included.
• What meta-data will you publish with each image?
  We will describe frames from sequences with rough categories like winter/summer.
  We will credit the original author and provide links to the original video and the user’s profile page. We might also extract and show other useful meta-data that is publicly available at that original source together with the images (e.g. date of the recording, etc.).
  All information relating to the images in the dataset are taken from public sources.
• Have you thought about data protection and privacy rights?
  Yes, absolutely! We will apply reasonable caution to prevent privacy infringements. The actual legal situation is quite diverse and specific to each country . This is why we need to publish the country information for each frame even for sequences where the author wants to remain anonymous. After release we will comply to complaints of individuals with reasonable cause who request the removal of individual frames that includes them, a relative, or private property.
• Why did I receive multiple request from your project?
  Please reply to any of our inquires . Send “No” if you are not interested in helping our project. Otherwise we are not able to distinguish between a rejection and having not reached you at all. We will try to contact you over different channels until we get a reply. Sorry if this generated unwanted traffic.
• Which accounts do you use for communication?
  These accounts are used by our team to contact authors:
  youtube channel “Dashcam Dataset”
  Google+/Email adress: dashcamdataset (at) gmail (dot) com
  Facebook: oliver.zendel.98

Thank you for your support. Please reply today and help us to improve road safety!