Technical referent of the computer vision department for the Rovereto office. Topics:Design and implementation of vision algorithms for detection and localization, with particular focus on 3D applications with laser profilometers and snapshot scanners. Calibration of 3D systems, both intrinsic and HandEye. Real-world testing of algorithms in harsh environments with high number of unknown noise variables. Applications with millimetric and sub-millimetric precision requirements. Optimizations of algorithms in order to reach high KPIs. Definition of vision system hardware following project requirements.

Topic: Acquisition of skills to develop Machine Learning applications and more generally Artificial Intelligence in Python, aimed at recognizing specific and predefined elements

DESIGN AND SIMULATION OF AN EMBEDDED APPLE DETECTION AND LOCALIZATION ALGORITHM FOR TETHERED UAV PLATFORMSDesign of a recognition system that can perform both detection and localization of apples on the trees. The algorithm, mounted on UAV platforms, can be used to find and pick apples in an automatic and time unconstrained way. The method uses a Faster R-CNN deep neural network architecture to detect fruits in optical images and stereo vision acquisitions to estimate the real world position of each apple. The algorithm also defines an ad-hoc decision strategy to robustly select the best fruits and remove unwanted predictions. The validation of the system is first performed on synthetic data created by modeling the scenario and performing simulations. Then the algorithm is tested on data from field acquisitions in local research orchards. The algorithm has also been tested and optimized on edge AI computing devices with promising inference performance.

Study of the different techniques for data analysis present in the literature, autonomous implementation of a Deep Neural Network from scratch using the PyTorch framework, validation of the results and evaluation of future improvements.