Computer Vision News - August 2016

Every month, Computer Vision News reviews a successful project. Our main purpose is to show how diverse image processing applications can be and how the different techniques can help to solve technical challenges and physical difficulties. This month we review a software for Detection of Incomplete Rows in Forestry , developed by RSIP Vision for one of its clients. Do you have a project in computer vision and image processing? Contact our consultants . Forest development is a domain with many practical applications in image processing and computer vision. One of our clients requested our help to provide a software to detect rows, in which development was incomplete . Computational forestry , supported by aerial images taken from planes or drones, analyses the status of the forest though 2 types of observations: in young forests, it is possible to detect individual trees; in more mature forests, canopies become connected, generating a continuum which precludes the identification of individual trees, therefore it is rows which need to be detected. In fact, using an average size of canopy, it is possible to estimate the approximate number of trees in a row . The present cases are a Brazilian eucalyptus forest at a stage of maturity of 4-7 years (close to the completion of the forest cycle) and a U.S. pine forest, where cycle lasts about 20-25 years and where canopies connect after 12- 15 years. Our software first identifies the rows, then it inspects them to find empty segments, as revealed by the absence of color clues which could be interpreted as a tree: green or, when working with NDVI , an index which is too low to indicate the presence of vegetation. Those void areas are called “ non-stocked areas ”. Our client wanted an additional optimization: he wanted to identify several contiguous rows with empty segments , in order to treat them as a non-stocked area, rather than individual missing trees or a non- stocked row. This problem involves two main challenges: the first one is detecting the rows, which are not necessarily continuous and do not necessarily respect a pattern. In general, rows are parallel, therefore we use an algorithmic approach which permits to identify an individual row and project it towards other rows. The second challenge is to detect spots which are empty of trees . NDVI images are of course more precise than RGB images, since the former enable to exclude trees showing a green external appearance but whose interior is dead and should not be counted in the forest inventory of trees. 24 Computer Vision News Project Project Detection of Incomplete Rows in Forestry