<p>Drones are increasingly utilized in surveillance applications, where object tracking and reidentification are crucial. This manuscript proposes a novel framework for single and multiple object tracking and reidentification in drone-based surveillance. The framework utilizes You Only Look Once Neural Architecture Search (YOLO-NAS) for object detection. YOLO-NAS incorporates quantization-aware blocks and selective quantization offering accurate detection. Further on, a confluence-based non-maximum suppression technique is applied to detect object in various occluded scenarios. The proposed framework includes a novel object tracker, the Densely Connected Bidirectional Long Short Term Memory tracker (DC-Bi-LSTM), which extracts spatial and visual features from YOLO-NAS. A novel occlusion handling reidentification mechanism is designed for single and multiple objects. The proposed framework is evaluated against state-of-the-art models using the VisDrone, UAV123, and UAVDT datasets. A comprehensive ablation study is conducted on the UAV123 dataset, demonstrating that the proposed framework outperforms other models, achieving a Precision of 97.19%, Recall of 97.8%, MOTA of 94.53%, Rel.ID of 9.26%, and F-score of 97.49%.</p>