The image processing algorithm for a 360-degree camera

Abstract

The goal of the thesis is to make an image of an entire row of fruits (in this case, tomatoes) in a greenhouse farm, which will be captured using a 360-degree camera. The picture produced will be utilized for various reasons, such as counting and monitoring. To begin, this thesis will review the basics of computer vision and introduce essential issues. The features of the 360-degree video, as well as their technological specs, will be discussed next. The OpenCV library may now be used to evaluate the data collected in the greenhouse in the form of 360-degree video. It is possible to begin video processing and picture stitching to get the desired outcome with all of that knowledge. However, the fisheye lens causes significant distortion, necessitating extra procedures to undistort the image using methods such as cube mapping. There are also flaws in determining the video's speed, which will result in an undesired outcome. This issue may be solved by using dynamic stitching, which calculates the movement speed in real-time. All of the above techniques have resulted in a few different algorithm implementations. An assessment utilizing a generated video with all the controlled parameters is used to quantify the mistakes caused by several variations of the algorithm in order to choose the optimum technique. The pixel differences technique delivers the best result with a decent speed after a lengthy testing period. Furthermore, future enhancements for best practices in picture capture and processing for this project will be offered.