Autonomous flying drone for vehicle classification

From ISLAB/CAISR
Revision as of 13:38, 18 October 2021 by Islab (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Title Autonomous flying drone for vehicle classification
Summary Building an autonomous flying drone for vehicle classification
Keywords Deep learning, robotics, drone, navigation, obstacle avoidance
TimeFrame VT2022
References
Prerequisites Deep learning, image analysis
Author
Supervisor Cristofer Englund, Fernando Alonso-Fernandez, Martin Torstensson
Level Master
Status Open


Building an autonomous flying drone for vehicle classification

Problem Vehicle detection and identification is one of the basic building blocks for many applications within traffic safety. Ground-based autonomous vehicles or land drones have difficulties in performing this task due mainly to the location of the sensors they carry. Land drones carrying cameras located at such low heights are not able to capture the necessary information for performing accurate classification algorithms. Solution The proposed solution takes cameras to a higher point of view (compared to a wheel-based ground vehicle) where images can collect a more complete and clearer picture of all objects in the environment. In the case of vehicle detection and classification, a flying drone could give more freedom to find good images of the vehicles in the environment. Moreover, a flying drone allows for faster movements which translates into shorter times of operation. It will be necessary to study the different options that exists in the market for ready-to-fly drones and other hardware and software. One of the options available in the market that could be considered for this project is the CrazyFly 2.1 drone (https://store.bitcraze.io/products/crazyflie-2-1) together with the Drone Autonomy Suite from CentEye (http://www.centeye.com/drones-and-robotics/perception-and-autonomy-suite/) On the software side; first, it will be necessary to collect as many images as possible from different vehicle manufacturers and their different car models in order to properly train a ML algorithm. Results should be presented including an accurate description of the validation methodology.