Deep Learning Based Target Tracking and Diving Algorithm in Kamikaze UAVs

This paper presents target tracking and target diving algorithms for Quadrotor-type Kamikaze UAVs (unmanned combat aerial vehicles, UCAVs). In the proposed approach, UAV's terminal attack dive analysis, aerodynamic features, and target dive physics flight are examined. A hybrid target tracking algorithm is developed using the Faster R-CNN ResNet101 model. The developed hybrid target tracking algorithm was observed to track targets with a 99.97% tracking accuracy and a frame rate of 156.89 FPS. In the Gazebo simulation environment, the UAV performs a dive maneuver at a constant speed in the x and z axes and a flight path angle of 53.13° to the targets it detects under dynamic environmental conditions.