Towards Integrated Image Contrast Models in Segmentation of Trees

Computer vision is an area in high demand which is bringing new trends for urban and rural applications. Some examples can be found in autonomous navigation projects, monitoring services, fruits/grain harvesting, pest control, and so forth. However, drastic or even unperceptive changes in the image acquisition process limit the development of these applications, especially for problems that require solutions for uncontrolled environments such as outdoor areas. Thus, the definition of what a machine is looking at is a challenging task. In this study, we dealt with the image segmentation problem in order to develop a method to delineate tree trunks, their branches, and foliage. As tree detection is a crucial topic in mobile robotics, we investigated it to give an initial interpretation of external scenes. We prepared an image dataset to validate the proposal in which two classes were defined, tree and non-tree. The pixels of each image were classified based on the proposed method, and the results show that our method obtained a positive result of 91% accuracy.