Misbah Bibi’s research while affiliated with Jeju National University and other places

Existing object detection and annotation methods in surveillance systems often suffer from inefficiencies due to manual labeling and a lack of accurate distance estimation, which limits their effectiveness in large-scale environments. These limitations reduce the speed and accuracy required for real-time surveillance, especially in scenarios that necessitate simultaneous monitoring of multiple feeds. To address these challenges, this paper proposes a framework for automated object detection and annotation, specifically designed for surveillance applications. The framework incorporates both manual and automatic modes, offering flexibility in object labeling. A synthetic data is created by using the blender tool which emulates the real-world security and surveillance environments, is utilized to train a yolo model for fast and accurate object recognition and identification. Moreover, for precise distance estimation a depth estimation model is used for calculating the distance for detected objects. The proposed architecture presents both manual and auto modes for the object detection by incorporating the both models in proposed framework. The auto mode of the proposed architecture increases the efficiency of the large surveillance and monitoring applications while lowering the manual labor. The model’s performance is evaluated with state-of-the-art models to assess its performance for auto detection and recognition. The precision and recall above 90% shows that our fine-tuned model demonstrates improved results on synthetic data. With real time surveillance and risk assessment as the foundation, this method seamlessly overthrows drawbacks of existing methods.

Accurate Object distance estimation with recognition is essential for various computer vision (CV) applications, including autonomous vehicles and many military operations. While significant advancements have been achieved in supervised and self-supervised techniques for short-range and real-time recognition, existing methods often focus on monocular depth estimation. They are constrained by the limitations of supervised deep learning (DL) models. Taking advantage of temporal information from sequential frames through attention mechanisms to address these challenges offers a promising avenue for enhancing recognition quality. For a real-time military object recognition system, this study integrates RGB images along Depth maps from the KITTI Dataset for short-range measurements. Due to the limitation of the KITTI dataset, a synthetic dataset is generated for long-range object recognition. YOLOv8 is trained for real-time object detection, utilizing the KITTI and a synthetic dataset for long-range analysis. Our method achieves an RMSE of 1.24 meters and an RMSE(log) of 0.18 for depth estimation, outperforming existing approaches. Furthermore, the system accurately detects objects at distances of up to 250 meters, with an average inference time of 15 ms per frame for short-range detection and 18 ms per frame for long-range detection. Comparative evaluations against state-of-the-art methods demonstrate our approach’s superior accuracy and efficiency.

Accurate depth information is crucial for various computer vision applications such as augmented reality, 3D modeling, and autonomous vehicles. Recent advancements have significantly improved both supervised and self-supervised methods for depth estimation. However, most current approaches primarily focus on monocular depth estimation and face challenges in overcoming quality limitations due to the inherent constraints of supervised learning in deep neural networks. Incorporating temporal information from sequential frames can enhance the quality of these methods. This paper explores innovative methods for integrating recurrent blocks into existing pipelines for supervised depth estimation using convolutional long short-term memory (convLSTM). By utilizing convLSTM, we can capture a wealth of valuable information. To accurately measure object distances, we employed a geospatial approach. After extensive analysis of training methods, new deep neural network architectures have been designed specifically for monocular video depth estimation. Our research emphasizes using an attention mechanism to extract information from previous frames. The proposed approach effectively measures both long and short-range object distances by combining geospatial and temporal mechanisms, demonstrating superior performance in monocular depth estimation tasks.

Accurate long-range object detection is essential for applications such as security and surveillance. However, existing datasets often lack the complexity needed to represent real-world outdoor environments, resulting in limited performance of object detection algorithms at extended distances.Synthetic data generation offers a way to address these limitations by creating varied and realistic training scenarios. To address these limitations, we propose a novel approach utilizing BlenderProc procedural generation and photorealistic rendering to create a synthetic dataset that captures diverse and realistic outdoor scenes for the objects detection. We trained YOLO model on this dataset and compared its performance with standard YOLO model. Our approach achieved a precision of 94% and recall of 96% for detecting objects at distances exceeding 120 meters, demonstrating significant improvements over existing methods. These findings underscore the potential of advanced synthetic data generation techniques to enhance long-range object detection and address critical challenges in surveillance, remote sensing, and autonomous systems.