With reduced drone cost, object tracking and localization algorithms are well studied for robotics research. In most cases, the scene is stationary and/or motion is smooth so that control system reference input varies less frequently: only at the sharp edges generated by path planning algorithm. However, using systems such as co-robots among human beings, in windy situations and/or around more sophisticated structures will require rapid localization, object detection and obstacle avoidance solutions due to continuously varying controller input and path plan. This research effort benchmarks fundamental object detecting and tracking techniques under highly dynamic disturbances. To perform these tests an experimental system was developed that could create rapid angular motion in a controlled manner. A sensor suite of a visual sensor and an IMU is assembled into a rig. It is mounted on tip of a 6-DOF multi-joint mechanism which consists of a 5-DOF robotic arm and a pendulum mounted to the end effector. While robotic arm is programmed to simulate aggressive drone motion (max. 60 degrees/sc.), pendulum generates an additional highly dynamic angular motion about another plane. This motion can be used to benchmark various types of tracking algorithms, allowing for a comparison in robustness with regards to different motions. The specific motion that will be explored in this research is rapid angular motion. A feature rich landmark is used to execute the experiments. Velocity, acceleration and viewing angle are varied continuously to benchmark basic search-based detection methods using gradient magnitude which is the fundamental step for advanced algorithms. Finally, results with different algorithm parameters are listed to compare robustness of the solutions. Overall system repeatability and precision are discussed with various plots of 2700 processed images.

This content is only available via PDF.
You do not currently have access to this content.