Robot Target Tracking Technology Based on Efficient Detection D7 with Gabor Filter

Nowadays, the robot target tracking technology has gained significant attention due to its applications in various fields, including surveillance, autonomous navigation, and human-robot interaction. However, the traditional tracking methods often struggle with occlusions and rapid target movements, highlighting the necessity for advanced. Hence, the existing Deep Learning (DL) model namely Convolutional Neural Networks (CNNs), have shown promise in feature extraction and pattern recognition, but the requirement for extensive labeled datasets, computational resource demands, and real-time processing constraints are remained as challenges. To overcome those issues, this research proposes an Efficient Detection D7 with Gabor Filter (EfficientDet D7-GF) by using CVPR 2020 MOTS challenge dataset. Then, the collected data is further preprocessed with Kalman Filter (KF) to reduce noise and min-max normalization to scale the data within specific range. After that, GF is utilizing to extract the features. Finally, EfficientDet D7 is employed to detect the object by using a compound scaling method to optimize backbone, feature network and box/class prediction networks. From the results, the proposed EfficientDet D7-GF achieved outperforming results in accuracy of 99.62% when compared to existing CNN developed by Yann LeCun (LeNet-5) model.