GoodVision Video Insights provides analytics of video footages from standard cameras placed lower above the ground as well as from flying drone cameras, as for each project, different approach - or its combination - might be utilized.
While standard surveys are suitable for traffic measurement on selected traffic places of importance (roads, junctions, pathways), drone footages are ideal for high-height videos (50-250m) overviewing a big scene, be it a large intersection, roundabout, superhighway or part of a town.
For each of the two types of footages, we have developed a special kind of processing to deliver the best output. GoodVision Drone High-Detail Processing is trained specifically for high altitude drone views with higher resolutions, giving accurate results on both top-down and tilt scenes. It is also the best processing option for videos from cameras fixed in big heights.
To achieve the accurate results, please make sure to follow these requirements.
1. Static viewpoint
GoodVision Video Insights can detect traffic objects from unstable or moving cameras normally but for the tracking purposes, it is necessary to have a stable fixed view. Moving camera would cause moved trajectories on the analytics page and make a proper analysis impossible.
That is easier to achieve by static cameras placed on posts, houses etc. and harder by a footage from a flying drone. We have introduced a stabilization procedure for drone processing which helps to "hold" the scene fixed for slight shaking and shivering of a drone but always try to avoid any unnecessary motion, rotation, zoom etc. when making a drone footage. The more stable the camera will be, the better results you' ll receive.
It is advisable to check your recording before uploading it to our system. If there is a moving part of the footage (like recording when flying, zooming or getting to the best tracking spot), consider removing the moved track from the video which you want to have processed. You would still get accurate results but it is much easier and more transparent to do the analysis when the trajectories are compact.
Also beware when uploading multiple drone videos to Cameras in GoodVision application. Video multiupload is possible both in one pack sequentially but if you intend to put more videos into one Camera (e.g. because of more drone take-offs for monitoring the same place, more chunks of videos etc.), try to make the single recordings as stable and consistent as possible. If everything goes well, the system will reward you with amazing performance like this:
However, it can easily happen that some parts of the video were not in focus all the time during the footage, as the drone would turn a bit or change its position. That influences the analysis negatively, as the number of vehicles in the "negative zones" would necessarily be lower than in reality, just because they were not in the camera view for some time.
For such cases, we implemented a mask overlaying the camera screenshot in the Video Insights application which lets you know, where these zones occurred. You will still see the objects' trajectories there and can still put lines and filters to measure traffic volumes and travel times on the grey zones but it' s not recommended because the outputs will not be complete by definition. You can consider the mask as guidance where you should be careful with the results and rather concentrate on the bright parts with complete data for your analysis.
Example of the mask overlay on the scene guiding you where not to place your lines or zones to assure the best accuracy
2. Object distance
For a successful drone processing, the footage can be taken both from UHD drone or fixed cameras in heights from 50 to 250 meters above the ground or from the monitored traffic objects. Meaning that on 250m high scenes with top-down or steep view, the objects would be this far and the results will be accurate.
Be careful with camera tilt on drone footages, if the camera is not angled directly downwards, object distance would be much greater for objects further away. Good Vision Video Insights can still deliver results for bigger distances but cannot guarantee a proper accuracy for objects too far away.
For this reason we recommend to prefer top-down drone footages but if you require a tilt scene for your analysis, try to record in lower heights than the maximal 250m. Top-down view is best for higher altitude footage and highway/road traffic focus, whilst angled camera on low altitude flights will allow for better detection of smaller objects like bicycles and motorcycles.
For drone cameras flying up to or around 50m height, please choose the standard video processing option which is more suitable for this type of input.
Camera resolution heavily influences the quality of the video and the quality of the computer vision processing. For footages taken in bigger heights 100-250m, the input resolution shall be UHD/4K (3840 × 2160/4096 × 3112) as this allows even the smaller objects to be recognized reliably from higher altitudes.
For videos from lower heights the footage can be of 2K (2560 x 1440), or FULL HD (1920 x 1080) resolution. GoodVision can provide declared accuracy level and detect objects of 20px size with drone HD processing on ultra-HD footage.
4. Frame rate
Frame rate of the video defines the fluency of the object’s motion and affects the tracking ability of the video analytic system. For Drone High-Detail Processing, ideal frame rate is 30 frames per second. Bigger FPS than 30 does not bring any serious impact to detection quality. On the other hand, FPS lower than 15 might cause tracking problems on various kinds of scenes. Our recommendation regarding frame rate for drone footages is 25-30 FPS, considering FPS 15 as a minimum for results to be delivered.
5. Traffic objects
Drone High-Detail processing is trained to bring accurate results about the ground traffic from up to 250m high altitudes like drone footages or high-rise building cameras. From these heights, it would be inaccurate to track pedestrians as smallest traffic objects through the scene. We’ ve decided to remove the pedestrian class from the outputs so data for pedestrians are not available for videos processed via drone processing. In return, you' ll get exact results for all other traffic classes on the scene. If you are interested in pedestrians for your drone footage, please adjust the height of the drone not to exceed 50m above the ground and use standard processing for your video.
6. Camera focus
The scene should be focused at the video start and no further focusing shall be applied. We have noticed on DJI footages that sometimes the drone tries to auto-focus during the recording which makes scenes to blur temporarily - we strongly recommend to turn off the autofocus while recording.
7. Camera hardware
GoodVision Video Insights accepts digital video footage from any type of camera with standard optics.
8. Other recommendations
- For recommendations regarding aspects of the scene (scene lighting, shadows, obstacles in view) or camera parameters (shutter speed, lens quality and distortion) please see standard camera processing guidelines.
- By drone footages, processing of time-lapse videos is limited and it is not advisable to use time-lapse function while making a drone footage.
- For video length, file type or file size we have no limitations. Minimum video duration for processing is 1 minute. Feel free to upload and process any type of footage adhering to the above mentioned guidelines.