F-RCLSTM: FASTER REGION-BASED CONVOLUTIONAL LONG SHORT-TERM MEMORY-BASED OBJECT DETECTION FOR ADVANCED UAV PATROLLING
Obad Abdullah Yousef Rabea, Lufei Xia, Al-Qubati Mohammed Ahmed, Berk B.A. Ba geri, Liang Wang · 2024
Unmanned aerial vehicle (UAV) patrolling application is developed using panorama stitching based on Bundle Adjustment and Improved Shape-Preserving Transform (BA-ISPT), simplified three-stage ghost elimination and Faster Region-based Convolutional Long Short-Term Memory (F-RCLSTM)-based object detection approach. BA-ISPT eliminates the parallax effects and achieves accurate alignment for obtaining the seamless panoramic image with genuineness. The three-stage ghost elimination process removes the ghosts using image information learning. Finally, the object detection is performed using F-RCLSTM which is formed by combining the Faster Region-based Convolutional neural networks (F-RCNN) and Long Short-Term Memory (LSTM). Experiments are performed using VisDrone 2019 dataset and the results showed that the proposed BA-ISPT and F-RCLSTM-based model improved the patrolling with high accuracy, low error rates and low complexity.