Is new, lightweight handheld suitable for woodland conservation and census? Can it enable more precise and specific tasks such as longitudinal inspection of individual plants & trees
The research team in Oxford Robotics Institute have developed a device and algorithms for mobile mapping. We are interested in applying this technology to woodland conservation
Aims of the Project
To explore the use of handheld sensing as an alternative to fixed LIDAR scanners
The research team in Oxford Robotics Institute have developed a device and algorithms for handheld LIDAR mapping. We are interested in applying this technology to woodland conservation. Currently Terrestrial Laser Scanning (TLS) is used in Earth sciences to reconstruct forests or physical structures – with survey of Wytham Woods by NPL in 2015 being a notable example. This procedure, while very precise, is slow and laborious.
By scanning with a mobile sensor, we can achieve scans at the speed of walking. In addition, we have also technology which uses deep learning to identify a descriptor (a kind of fingerprint) for each tree enabling re-detection and positioning of individual trees. This can enable re-location within the forest and in doing so can enable automatic individual trees measurement - for example relating the individual back to its id in the original census database enabling growth measurements, forest health survey and canopy density measurement – for example quantitatively measuring the rate by which a forest recovers from a forest fire or pollution incident.
Specific scientific challenges are how well spot measurements match those made manually by traditional means and how well one can correlate indicative measurements to whole-forest properties such as CO2 absorption. This involves the cross-disciplinary application of technology originally developed for autonomous robotics and self-driving cars to an entirely different domain.
Overview of the device (used to map New College, Oxford):
Overview of experiments in New College:
SLAM – Simultaneous Localisation and Mapping (One constituent technology):
Deep Learning for Localisation in Forests:
Existing 3D model of Wytham Woods (from NPL,
Methods to be used
3D motion localisation and mapping (known as SLAM) as well as neural networks for individual tree detection
Specialised skills required
Probabilistic Estimation and Deep Learning
Computational Vegetation Ecology
Please contact Maurice Fallon on firstname.lastname@example.org if you are interested in this project