Enlaps is a company that offers plug&play Tikee timelapse solutions (videos accelerated over a long period of time). Their solution is composed of a Tikee camera equipped with two wide-angle lenses and a solar sensor, allowing it to create panoramic time-lapse photography in complete autonomy. Thanks to Wifi or 4G communication, the images are sent to their MyTikee web platform where they can be retrieved, edited and exported as videos.
The final images produced by their Tikee box come from two wide-angle cameras. It is therefore necessary to merge the images captured by the two cameras and smooth up this fusion.
Neovision has developed algorithms to merge image pairs to obtain a panoramic image. This panorama problem, very classic in photography, becomes much more complex in the case of a video. This solution allows the automatic merging of images, without the merging area being visible on the final video.
The Tikee box from Enlaps is now able to produce very high quality panoramic timelapse videos automatically. The merging of the images is directly managed by their software solution, without the need for the user to manually intervene on the processes. Neovision’s technology enables Enlaps to deliver on its promise by offering a true plug&play solution!
When Enlaps arrived at the Tarmac, Neovision was still housed there. The two startups were therefore neighbors. Thanks to the proximity, the two teams got to know each other which quickly led to obvious areas of collaboration. While the problem of image fusion is commonplace in photography, applying this technique to videos is much more difficult. Faced with this challenge, Enlaps turned to a young but already recognized expert in artificial intelligence: Neovision. With numerous successful projects in image processing and analysis, Neovision had the necessary knowledge and skills to partner with Enlaps.
This project was carried out in several stages. The first step was to merge the images from the two cameras. Neovision then relied on its network and consulted the opinions of several researchers. At the same time, the team was conducting technical research. This research aimed to establish a comprehensive state of the art of the best image fusion techniques. Following this, Neovision produced a prototype to test the results of its research. The goal was to merge a first set of hundreds of image pairs.
The prototype was satisfying so Neovision optimized it. This step was designed to ensure that there was no visible gap between the two images. The prototype was working so we started the industrialization stage! In collaboration with ENLAPS, Neovision defined the architecture and communication protocols for the Docker Images. The latter supports the calculation of fusion parameters before merging the two images.
The challenge was met! However, there were still many areas for improvement as far as the rendering of the video was concerned. The first one was to widen the fields of view (FOV). That is why Neovision tried to increase FOV by region of interest and masks. The result was a gain of 21% in width against a loss of only 2% in height. This is what we call panoramic vision!
This achievement demonstrated a beautiful collaboration between Neovision and Enlaps. The Tikee solution from Enlaps has been on the market for 2 years, and today there are 3500 Tikees in over 50 countries. The quality of the panoramic images of the cameras is unanimously appreciated by Enlaps’ customers. We will let you be the judge of the quality of our work with the video below.
Finally, following this first successful collaboration, Enlaps and Neovision have been working together to develop new AI functionalities that open up new B2B markets for Enlaps. But what are they?
Producing beautiful images has many advantages. One of these advantages is to be able to accurately see many objects in timelapse scenes. Among these “objects” are motorized vehicles and people. Neovision therefore developed AI technology that detects and recognizes these two types of “objects”. That is why the Tikee solution from Enlaps enables the analysis of visitor numbers, attendance or occupancy of certain public places (parking lots, shopping centers, etc.) and provides valuable information to their business customers.
IMAGE PROCESSING, DOCKER, C++, MACHINE LEARNING, OPTIMIZATION
16 October 2020
Computer Vision, Machine learning