Choosing the site
From the early stages of the project, in the beginning of March 2010, our top priority was to find the location from which we could capture the most spectacular images of the city in order to reveal all its splendour and beauty. Sevilla is a city with only a few high buildings, which are well located and adequate to carry out our project. The only place with a considerable height, 150 meters high, is the Alamillo Bridge, but as it is located in the Northern region of the city, it only allowed us to have a excessive far view and did not meet our expectations.
The second highest point in the city is La Giralda, which is 98 meters high. Its bell tower is located at the heart if the city center, so a priori it seemed to be the best place to work, but it had two main disadvantages: Firstly, the Guadalquivir river is hardly seen from La Giralda, and secondly if the photograph was taken from the Cathedral's tower, this emblematic monument would not be present in the photograph; the truth is that a panorama of Seville without two of the most important icons would lose much of its essence.
Finally, the chosen site was the Torre Schindler, an observation tower which is over 60 meters high and was built for the world exhibition Expo'92. This high tower in the banks of Guadalquivir river proved to be ideal for our project, since the entire center of population lays within a 5 km radius from the tower. From this site, we would have privileged views of both the city and the river.
The Torre Schindler has two panoramic lifts and a high-capacity lift, but unfortunately these were out of order when the project was carried out as they were being refurnished; therefore, we had to climb the stairs up to the 16th floor, in which the upper terrace is located, with all the photographic equipment (we could not be lucky in all aspects). There was a solid railing which allowed us to mount and install the photographic devices to obtain a wide horizontal vision field (about 290 grades). It was exactly what we needed.
A project of such a big magnitude as Sevilla 111 Megapixels is only possible if some kind of robotized device is used to automate the process as much as possible. After analysing the different devices available in the market, we were not satisfied with their features, especially because after the first testings we suspected that we had to be very imaginative when controlling the camera. Eventually, we decided to build our own photographic robot, which could be adapted to our changing requirements and offered us a total control over all factors.
Finally, the "D-Day" arrived in early April. Everything was at last ready to start the project. Once the material was lifted and prepared, we discovered our first obstacle: we found out that our design had a fatal error. The control electronics was built into the robot, so the display and buttons were completely unreachable, when the camera was tilted to either end. The robot was installed in the camera bracket which protruded about 20 cm from the tower railing. This first day, we decided to place the bracket on a camera tripod not to leave empty handed and we made a testing of about 2000 pictures, covering about 160 grades in the horizon.
Another disturbing discovering was the fact the wind gusts were so strong that they were able to blow the coffee inside our cups. When the images were displayed in the computer, we discovered that most of them were blurred even though they were shot at 1/2000. In order to solve this problem, we had to build a completely new bracket with separate electronics and redesign the structure to make it much thinner and offer less resistance to wind.
Once the bracket was ready, we tried to reach our production goal, which was about 7000 photographs. Our shutter speed was one image every 4 seconds, which plus the focus time required for each line, the replacement of cards and batteries, resulted in an actual speed of 4.8 seconds per image. In every session, we worked 10 hours non-stop to shoot all the images.
Seville has a very good natural illumination in May until 8 pm, so we did not have to start very early; we could start working at 10 am. After several attempts, despite the weather and wind forecasts, we did not manage to capture all the images we required to obtain an optimum throughput. The problem was sometimes the sun which hid behind the clouds or the sudden rainfalls, but above all the main problem was the wind gusts which appeared suddenly sometimes for seconds and other times for a long time, which forced us to stop the process.
At the end of June, we made a last attempt but the wind did not let us be successful either. The environment temperature already reached 35 degrees Celsius and the heat was visible floating in the horizon, so we had to postpone the project until the end of the summer.
Changing the strategy
After the summer, things changed in several aspects. The first issue was that the number of gigapixel panoramas was increasing and the record was in 70 gigas, so we changed our target to break the barrier of 100 gigapixels. This meant that we could not shoot all the pictures in just one day, we had to control the wind and we needed to monitor the images in real time to correct and repeat the possible faulty images. The solution was to add an anemometer to our robot, so that we could be able to control the wind speed at all times. Now, when the wind exceeded the 9 Km/h, the device would stop shooting automatically.
Moreover, we also added a remote control software to the robot, which allowed us to control all the devices from a remote computer. Being the camera and the robot connected to the PC, we could control the robot's operation and monitor the images as they were being captured, thus controlling their quality in real time. We were even able to focus the camera without touching it, so we decided to add a second computer which controlled the first computer through a remote desktop. Thanks to this, we could place the control computer in the same tower, 4 floors below, far from the heat and with an environmental luminosity which allowed to review the images directly on the monitor. The system stopped automatically when the winter speed was over 9 km/h, so we simply needed to wait until the wind abated. When the wind speed was adequate, we only needed a mouse click to resume operation.
One of the issues was that some people and cars appeared repeatedly or cut in different images. As we were monitoring in real time, we could reverse the operation and repeat all the necessary photographs at any time. When the end of a line was reached, we focused again by using the camera's direct vision with real-time images and 60 meters away from the camera.
At the end of September, the temperature was at last 32 degrees Celsius, which was enough to maintain the quality of the images. We just needed 4 days with the same weather conditions so that the light was the same in all the shootings. Unfortunately, we were forced to say goodbye to our original idea of capturing cotton clouds which would have created a very nice view of the panorama.
On September 27th and 28th, we started to shoot from the horizon line and below. The first day we took pictures of the lower half and the next day of the upper half. We worked both days at the same time of the day, so that the light angle and luminosity were identical. The result was perfect and in fact during the processing of the images, we were not able to distinguish between the photographs of the two days.
As we stayed all day in the tower, we did not have much time to make backups at night. We spent the third day reviewing all the images to locate possible faults or errors. In this review, we discovered that a photograph of the center of the panorama was completely black. The camera shot but the image was neither in the raw file nor in the jpg file. Luckily, the solution was simple: We had to take that photograph on the next day at exactly the same time. Our custom camera bracket was always mounted in the same location, so the robot allowed us to repeat any given photograph by simply using its horizontal and vertical coordinates.
At last, on September 30th, after 6 months of hard work and 12 attempts, we had enough images to make our project come true.
The last days we took about 14,000 photographs, which included all the additional images we had to repeat to correct the errors caused by vehicles and pedestrians in motion. Since the beginning in April, we took over 35,000 photographs. The camera is a Canon 5D mkII with an 400-mm lens with duplicator also by Canon, which is equivalent to 800 mm of focal distance. The shutter speed was 1/800 S with an aperture of f16 and ISO800. These settings worked well from the clearest areas to the darkest areas, despite the contrast difference between them is very high.
Working with a 800 mm lense is fairly difficult because it is extremely sensible to any motion. The video shows how the camera with a 800 mm lense moves with a 11 km/h wind. Since the Guadalquivir river lies in a valley, there is always wind on top of the tower, caused by a tunnel effect, even though the wind is calm at ground level. This is the reason why we had to work with the mirror up and the anemometer connected to the robot.
To edit the definite 9,750 images we basically needed two things: on the one hand, a powerful computer and on the other hand an editing software which is able to make the most of the computer power. The chosen computer was a PC with two 6-core Xeon processors, 40-gigabyte RAM and 8-terabyte HDD. We chose the renowned Autopano Giga software which had proved to be a very powerful tool in similar projects.
The first step was arranging and sorting all the images until we located all the 65 rows consisting of 150 images each. We had a jpg version of each file, apart from the version in raw format, so we could view and sort the images much quicker. We also had to discard all the repeated images, the images with incomplete elements, such as parts of people and cars, etc. After a long and tough week devoted to copying, reviewing and sorting the images, we managed to select the definite 9,750 images.
Then, the images were developed to obtain the most faithful color. The light in Spain is very bright and produces images with a great contrast and vibrant colors, so we only had to edit the photographs which were taken late in the afternoon, because even though they had a warm golden effect, they did not match the rest.
When we first tried to load the images in Autopano Giga software, the application was completely blocked. After several testings we discovered that the version 2.0 of the program has an important limitation; it was blocked if more than 5,000 images were used, which was a big problem for us. Fortunately, Autopano Giga Beta version 2.5 was released just in time and solved not only this problem but many others which would appear afterwards. This new version included a filter which removed the haze effect that worked very well with the images close to the horizon. We used this filter with a total of 50 images in our final panorama.
The good news is that this new version was capable of loading all the images in just 1,5 hour and processing the panorama in 10 hours to reach the required .pano file; the bad news is that the file required a further extensive editing work to correct errors and settle all elements down in the proper place. This was for sure the beginning of our worst nightmare. Firstly, the program stuck once in a while, and simple operations, such as a mouse click, a digit change or tool selection, generated a delay between 3 and 20 seconds; thus, any operation on the program was simply distressing. Fortunately, just when we were about to throw the towel for good, another new version of the software was released which solved almost all our problems, preventing the program to be stuck and improving the edition speed dramatically (now, delays lasted only 1 and 5 seconds). Once again we won the battle and after two weeks of infinite patience, the work was almost finished. We only needed to perform the final rendering.
In mid November we started the final rendering of 3 files in total which took about 100 hours, that is, over 32 hours each. The rendering took a whole week, because we had to turn off the computer, clean the memory, and reload the files to render the three files separately. Finally, we only had to edit the resulting images in psb format, correct the rendering errors and blur people's faces and car plates to avoid identification.
We chose the Krpano software as a viewer, which works with Flash and therefore it is compatible with most Web browsers and most users have the Flash plug-in already installed. This was the strongest point of this viewer, since nothing needed to be downloaded to see the panorama. This software also supports several interactive operations and can be comprehensively customized to our requirements. We were quite amazed about the operation of the program and its possible capabilities. For the implementation we used our own design team, who designed the Web, our logo and all the elements necessary to launch the panorama in Internet.
The resulting panorama consists of 140,000 small images which are displayed on the browser as required. We can offer the finest details of all the images on the monitor with an extremely low data transfer to enable most users to enjoy this experience as much as possible.
This project with an unquestionable cultural and touristic interest would not have been possible without the sponshorship of the online shopwww.SuperInventos.com, which apart from sponsoring the production of the project, has cooperated in all the stages of the work.