Traffic data collection studies using drones have proven to be very useful given their ability to record still images and video from above rather than below. Indeed, the broader perspective provided by having a single camera in the air can be far more efficient than deploying multiple land based cameras capturing the same area. This comparison holds true for areas where a single drone can be flown within eyesight but what about much larger areas where it can't?

Case in point is a recent study undertaken by QC of Belle Isle located on the northern border of Detroit, MI in the Detroit River between Canada and the United States. Belle Isle is connected to Detroit by a bridge and covers an area of approximately 1,600 acres including a state park visited by thousands of tourists and local residents each year. Local authorities wanted to know  the impact of traffic at various times of the week and also measure parking capacity on an hourly basis.

Capturing this amount of data using traditional land cameras and on foot, while possible, was projected to be beyond budget and further challenging to coordinate within the timeline of hourly measurements. Much the same issues existed for drones as well in that the area was simply too large for a single drone to cover hourly. Moreover, coordinating multiple drones working in synchrony would be both too expensive and difficult to deploy. In short, how could QC cover this much area on an hourly basis and provide the needed data?

‍

Quality Counts worked with Lee Hinshaw of Kestrel Imaging to come up with a somewhat new and novel approach. The solution was to combine a fixed wing aircraft with mapping software originally developed for drones. It works something like this: A DSLR camera is mounted inside a Cessna 172 airplane modified to accommodate a camera in the belly so the camera is pointed straight down or 90 degrees from the ground. The airplane then climbs to an altitude of 1,500 feet and makes several passes over the island, in this case seven, while the camera records an image once every 2 seconds. This process yields just under 300 high resolution images which are later "stitched" together to form a single high resolution map of the entire island. In the case of Belle Isle, each mission described here took about 20 minutes to complete - well within the timeframe needed to create data on an hourly basis.

The software used to create the flight plan for the airplane was developed out of San Francisco from a company called DroneDeploy. Again, this software was designed to be used by drones where a user enters the make and model of the drone used, the intended altitude (this is limited to 400 feet under FAA guidelines) and of course the area to be covered. This information is then downloaded to the drone which then takes off, traverses the entire area in a grid like pattern while recording images every 2 seconds or so and then lands - all of this done automatically with no control required from the drone pilot unless needed to avert an emergency condition. Note that with both a drone or DSLR camera in an airplane, exact GPS coordinate information is stored with each image. This information is required in order to stitch all of the mages with precision. The image below shows the passes made over Belle Isle in the airplane with each blue dot corresponding to the images acquired.

‍

The last step in this process is uploading over 11,000 images to DroneDeploy for processing or "stitching". Each mission included just under 300 images and once processed, maps can be exported at various resolutions. In this case, the final resolution of exported maps was 8 inches per pixel. That translates into a JPG image about 90 megabytes in size with a resolution of about 30,000 x 15,000 pixels. This level of resolution renders an incredible amount of detail making the task of counting cars and other details quite easy. Additional image processing was added to the final map images by Kestrel Imaging to provide added sharpness and contrast. Finally, the edited map images were indexed and uploaded to a cloud account for QC processing and client delivery.

‍

Click here for an example of the finished map:
‍ https://qualitycounts-my.sharepoint.com/:i:/g/personal/mshields_qualitycounts_net/EeDrpoYxoXFGhLaM7gNfjywBy29W0S7BUyWBYTQQ_4whXQ?e=u9OlrN
‍(Click three dots in top left > View original to see image at full resolution)

‍

Capturing areas with an airplane is not without potential problems or concerns including weather, air traffic and proximity to a nearby airport. It's truly remarkable that all 38 flight missions over Belle Isle over 5 days were not interrupted once by adverse weather. And while an airplane can still capture images in light rain while a drone can't, thunderstorms are always a concern especially in the summer months. Close proximity to a nearby airport is also an important factor. Without a convenient place to land, costs rise sharply for an airplane forced to continue flying until the next mission burning fuel and money. Fortunately for this project, Detroit's first airport, Coleman Intl (built in 1927 and later replaced by Detroit International) was less than 3 miles away taking only 3 minutes or so to get over Belle Isle. Air traffic can be a problem as well but this of course is coordinated by air traffic controllers constantly in contact. Belle Isle is only a few miles from Windsor Airport in Canada restricting altitude to 2,000 feet but for this project, 1,600 feet was well within the limit. Finally, there are occasional "TFR" or Temporary Flight Restrictions placed on airplanes for a variety of reasons. For example, it is illegal to fly over Belle Isle under 3,000 feet when an event is in progress at nearby Detroit Field stadium. Fortunately, there were none during our project.

‍

An important aspect of this project was the preliminary testing done prior to its execution. Test flights were needed to determine the optimum altitude and airspeed needed to create reliable maps. As described above, the calculations afforded by the mapping software were built specifically for various drone models and not airplanes. Our own calculations combined with real world testing rendered far more confidence than simply "winging it" and proved to be both reliable and successful. The information and experience gained on this project will be equally valuable for future projects like this.

‍