Tree detection and geolocalization with depth-camera and deep learning algorithms

A new approach of low-cost terrestrial mapping sensors is the ZED 2i stereo camera, which has been introduced into forestry settings. A stereo camera simulates human vision, capturing two simultaneous images that can be used to estimate depth and motion of detected objects. This includes training the necessary detection software to run with the ZED 2i Stereo Camera, collecting data in varying environments, converting the local coordinates of detected trees to a global coordinate system, identifying the more reliable observations, and consolidating clusters of observations for same trees into a singular tree position. Studies were conducted on the overall performance of the stereo camera compared to extracted control data from drone orthophotos, and saw that more observations were detected in a forestry plantation when foliage had developed in May versus February tests when there were no leaves on the trees. Testing also compared data results from a tree plantation to a more complex, natural forest environment, which saw a reduction in overall observation retention after being put through accuracy filtering based on detection confidence and distance of detections from the camera. In brief, data was used to configure processing procedures to allow for the extraction of meaningful results (consolidated tree positions, comparison of different test path types and different test environments), which allowed for an opening assessment of the performance of the processing methods established against control data gathered in the same environments using established methods such as GNSS positioning or drone data extraction. Forestry processes are calling for the implementation of new technologies for automatic detection, and automated vehicles and machinery. This thesis lays the groundwork for future improvements in data interpretation and accuracy based on the developed procedures, highlighted assumptions, limitations, and next steps for working to implement the use of a ZED 2i Stereo Camera in the forestry industry.