School of Automation Science and Engineering

With the proliferation of mobile computing devices, the demand for continuous network connectivity regardless of physical location has spurred interest in the use of mobile ad hoc networks. Since Transmission Control Protocol (TCP) is the standard network protocol for communication in the internet, any wireless network with Internet service need to be compatible with TCP. TCP is tuned to perform well in traditional wired networks, where packet losses occur mostly because of congestion. However, TCP connections in Ad-hoc mobile networks are plagued by problems such as high bit error rates, frequent route changes, multi-path routing and temporary network partitions. The throughput of TCP over such connection is not satisfactory, because TCP misinterprets the packet loss or delay as congestion and invokes congestion control and avoidance algorithm. In this research, the performance of TCP in Ad-hoc mobile network with high Bit Error rate (BER) and mobility is studied and investigated. Simulation model is implemented and experiments are performed using the Network Simulatior-2 (NS-2).

the varying number of traffic sources, number of nodes, host velocity, and data sending rate. These two protocols are simulated and compared under different mobility models namely Random Waypoint Mobility Model, Manhattan Grid Mobility Model and Reference Point Group Mobility Model. The performance metrics that are used to evaluate the performance of the routing protocols are Packet delivery ratio (PDR), Average latency of data packets, Network control overhead (NCO) and Throughput.

Efficient routing mechanism is a challenging issue for group oriented computing in Mobile Ad Hoc Networks (MANETs). The ability of MANETs to support adequate Quality of Service (QoS) for group communication is limited by the ability of the underlying ad-hoc routing protocols to provide consistent behavior despite the dynamic properties of mobile computing devices. In MANET QoS requirements can be quantified in terms of Packet Delivery Ratio (PDR), Data Latency, Packet Loss Probability, Routing Overhead, Medium Access Control (MAC) Overhead and Data Throughput etc. This paper presents an in-depth study of one-to-many and many-tomany communications in MANETs and provides a comparative performance evaluation of unicast and broadcast routing protocols. Dynamic Source Routing protocol (DSR) is used as unicast protocol and BCAST is used to represent broadcast protocol. The performance differentials are analyzed using ns-2 network simulator varying multicast group size (number of data senders and data receivers). Both protocols are simulated with identical traffic loads and mobility models. Simulation result shows that BCAST performs better than DSR in most cases.

Realistic human action recognition has been emerging as a challenging research topic due to the difficulties of representing different human actions in diverse realistic scenes. In the bag-of-features model, human actions are generally represented with the distribution of local features derived from the keypoints of action videos. Various local features have been proposed to characterize those key points. However, the important structural information among the key points has not been well investigated yet. In this paper, we propose to characterize such structure information with shape context. Therefore, each keypoint is characterized with both its local visual attributes and its global structural context contributed by other keypoints. The bag-of-features model is utilized for representing each human action and SVM is employed to perform human action recognition. Experimental results on the challenging YouTube dataset and HOHA-2 dataset demonstrate that our proposed approach accounting for structural information is more effective in representing realistic human actions. In addition, we also investigate the impact of choosing different local features such as SIFT, HOG, and HOF descriptors in human action representation. It is observed that dense keypoints can better exploit the advantages of our proposed approach.

Binarization processing is a key step in most vein pattern recognition, but may create flaws such as leakage of some feature information. In order to overcome this problem, a new approach is explored in this paper which can extract topographic features directly from gray-scale vein images without binarization processing. First, an extreme graph of directional fields (EGDF) is calculated according to neighborhood information of every pixel, and then the intravenous main ridge of the image can be separated step by step from EGDF with the defined constraint template sets. After that, the sub-ridge is extracted from end points of the main ridge in range constraint image with the tracing method of maximal neighboring point. According to contrast experiment the proposed method is better than the traditional processing (performing binarization first and then skeletonization), in that it can overcome the problem that some separate vein lines are extracted incorrectly, and it can avoid extracting two neighboring vein lines as one ridge. As such, the proposed method is an effective way to extract topographic features for vein recognition

Self-supervised Multi-view stereo (MVS) with a pretext task of image reconstruction has achieved significant progress recently. However, previous methods are built upon intuitions, lacking comprehensive explanations about the effectiveness of the pretext task in self-supervised MVS. To this end, we propose to estimate epistemic uncertainty in self-supervised MVS, accounting for what the model ignores. Specially, the limitations can be categorized into two types: ambiguious supervision in foreground and invalid supervision in background. To address these issues, we propose a novel Uncertainty reduction Multi-view Stereo (UMVS) framework for self-supervised learning. To alleviate ambiguous supervision in foreground, we involve extra correspondence prior with a flow-depth consistency loss. The dense 2D correspondence of optical flows is used to regularize the 3D stereo correspondence in MVS. To handle the invalid supervision in background, we use Monte-Carlo Dropout to acquire the unce...

The core of Multi-view Stereo(MVS) is the matching process among reference and source pixels. Cost aggregation plays a significant role in this process, while previous methods focus on handling it via CNNs. This may inherit the natural limitation of CNNs that fail to discriminate repetitive or incorrect matches due to limited local receptive fields. To handle the issue, we aim to involve Transformer into cost aggregation. However, another problem may occur due to the quadratically growing computational complexity caused by Transformer, resulting in memory overflow and inference latency. In this paper, we overcome these limits with an efficient Transformer-based cost aggregation network, namely CostFormer. The Residual Depth-Aware Cost Transformer(RDACT) is proposed to aggregate long-range features on cost volume via self-attention mechanisms along the depth and spatial dimensions. Furthermore, Residual Regression Transformer(RRT) is proposed to enhance spatial attention. The propose...

Robot is a device, which performs human task or behave like a human-being. It needs expertise skills and complex programming to design. For designing a fire fighter robot, many sensors and motors were used. User firstly send robot to an affected area, to get live image of the field with the help of mobile camera via Wi-Fi using IP camera application to laptop. If any signs of fire shown in image, user direct robot in that particular direction for confirmation. Fire sensor and temperature sensor detects and measures the reading, after confirmation robot sprinkle water on affected field. During extinguish process if any obstacle comes in between the prototype and the affected area the ultrasonic sensor detects the obstacle, in response the robotic arm moves to pick and place that obstacle to another location for clearing the path. Meanwhile if any poisonous gas is present, the gas sensor detects and indicates by making alarm.

This study studies the protocol of robust synchronisation of second-order multi-agent systems via pinning control. First, it is shown that, for the second-order multi-agent system with disturbances, by pinning one node with small coupling strength, the whole system can reach synchronisation. Then, these results are applied to the cooperative control of multiple 3-degree of freedom helicopters systems with disturbances. Finally, simulations are performed to illustrate the effectiveness of the theoretical results.

—Poor illumination, less background contrast and blurring effects makes the medical, satellite and camera images difficult to visualize. Image fusion plays the vital role to enhance image quality by resolving the above issues and reducing the image quantity. The combination of spatial and spectral technique Discrete Wavelet Transform and Principal Component Analysis (DWT-PCA) decrease processing time and reduce number of dimensions but down sampling causes lack of shift invariance that results in poor quality final fused image. At first this work uses combined median and average filter that eliminates noise in the image which is caused by illumination, camera circuitry and sensor at preprocessing stage. Then, hybrid Stationary Wavelet Transform and Principal Component Analysis (SWT-PCA) technique is implemented to increase output image accuracy by eliminating down sampling and is not influenced by artifacts and blurring effects. Further, it can overcome the trade-off of Heisenberg's uncertainty principle by improving accuracy in both domains, time (spatial) as well as frequency (spectral). The proposed combined median and average filter with hybrid SWT-PCA algorithm measures quality parameters, such as peak signal to noise ratio (PSNR), mean squared error (MSE) and normalized cross correlation (NCC) and improved results depict the superiority of the algorithm than existing techniques. Keywords—Image fusion; Heisenberg's uncertainty principle; combined median and average filter; Haar wavelet; Stationary Wavelet Transform and Principal Component Analysis (SWT-PCA)

In the ever-evolving landscape of semiconductor manufacturing, ensuring the quality andreliability of Integrated Circuit Quad Flat Package (IC QFP) components is paramount. This thesis presents a novel approach to IC QFP defect inspection by leveraging the power of deep learning methods, specifically the YOLOv7 network. The primary objective of this research is to enhance the efficiency and accuracy of defect detection in IC QFP components while addressing the challenges inherent to the manufacturing process. The core contribution of this work is the development and implementation of a defect detection system based on the YOLOv7 deep learning architecture. YOLOv7's real-time object detection capabilities make it particularly well-suited for inspecting IC QFP components, where defects can vary in size, shape, and location. To facilitate model training and evaluation, a comprehensive dataset of IC QFP components with various defects has been collected and meticulously annotated. This dataset serves as a valuable resource for future research in the field of defect inspection. A significant portion of the research effort is dedicated to fine-tuning andoptimizing the YOLOv7 model for IC QFP defect detection. This involves experimenting with hyperparameters, data augmentation techniques, and model architectures to achieve state-of-the-art performance. To assess the effectiveness of the proposed YOLOv7-based method, a quantitativecomparison is conducted against existing defect inspection methods commonly used in the semiconductor industry. Performance metrics such as precision, recall, F1-score, and mean average precision (mAP) are rigorously evaluated, demonstrating the superiority of the proposed approach. The research emphasizes the practical applicability of the developed defect inspection algorithm in real-world manufacturing environments. Considerations such as speed, accuracy, and adaptability to varying production conditions are thoroughly examined.This thesis also outlines potential future directions for enhancing IC QFP defect inspection, including the integration of additional sensor modalities, the incorporation of anomaly detection techniques, and the exploration of edge computing for real-time inspection. In conclusion, this research presents a robust and efficient defect inspection algorithm for IC QFP components, leveraging the YOLOv7 deep learning network.