This senior thesis develops a real-time handwritten digit identification system using a Raspberry Pi 3B+ with a camera module, leveraging a lightweight CNN optimized with MNIST. The project highlights the effective im...
详细信息
ISBN:
(数字)9798350350821
ISBN:
(纸本)9798350350838
This senior thesis develops a real-time handwritten digit identification system using a Raspberry Pi 3B+ with a camera module, leveraging a lightweight CNN optimized with MNIST. The project highlights the effective implementation of deep learning on edge computing devices through seamless integration of CNN, TensorFlow Lite, and OpenCV's real-time image processing. The system is both cost-effective and precise, enabling real-time digit recognition tasks. This proposed work illustrates the potential of AI applications in education, industry, and commerce, setting the stage for future advancements in embedded AI systems.
In the context of increasing number of public threats, the development of automated real-time detection systems, capable of identifying crime activity and potential attacks in video recordings, has become a major prio...
详细信息
ISBN:
(纸本)9798331539986;9798331539979
In the context of increasing number of public threats, the development of automated real-time detection systems, capable of identifying crime activity and potential attacks in video recordings, has become a major priority. We propose accurate and efficient deep learning solutions for analyzing surveillance camera footage and signaling situations of imminent threat, based on detecting specific elements, such as theft masks and various types of weapons. We showcase the enhancement of YOLOv8 model with an attention mechanism, based on Squeeze-and-Excitation (SE), and with deformable convolution layers. The model is trained on custom datasets with optimization algorithms like Adam, AdamW and SGD. The performance tests are employed on a separate dataset, analyzing various specific metrics. Furthermore, a comparative analysis with alternative deep learning models, such as RetinaNet and Faster R-CNN, highlights the advantages and disadvantages of each model, providing insights into their performance under various conditions. Additionally, a voting mechanism is implemented to manage discrepancies in detections from multiple models. Extensive experimental results demonstrate significant improvements in detection precision and efficiency, making the proposed enhancements a viable approach for more accurate threat detection in real-time surveillance applications, while demonstrating our model's superiority in balancing detection speed and precision.
This paper presents a novel approach to modeling stochastic systems using Dynamic Probabilistic Automata (DPA), which integrates deterministic and stochastic elements within a unified framework. Traditional methods, s...
详细信息
In this modern era, the popularity of autonomous systems has increased manifold because they are replacing people in various jobs and are expected to become the backbone of modern society. Specifically, self-driving v...
详细信息
ISBN:
(纸本)9781665460873
In this modern era, the popularity of autonomous systems has increased manifold because they are replacing people in various jobs and are expected to become the backbone of modern society. Specifically, self-driving vehicles and robots are gaining popularity in a wide range of applications. However, traffic sign detection, a critical component of intelligent transportation systems, remains quite challenging since it needs to be done quickly, with high precision, and with high dependability. A fast, real-time, robust automatic traffic sign detection and recognition can support and relieve the driver and significantly increase driving safety and comfort. Out of many algorithms and frameworks available for traffic sign identification, i.e., object detection, one of the most popular ones is You Only Look Once (YOLO) since it provides accurate results with minimal background errors in most real-time processing tasks and has excellent learning capabilities. Motivated by this, this research article provides a novel framework, CLEAR, for traffic sign identification in adverse climate conditions using YOLOv5. We compare different models' speed, accuracy, and other metrics on a Traffic Sign Dataset obtained from the Open Image Dataset V6. Experimental results demonstrated that the proposed CLEAR model achieved the best performance with a Mean Average Precision of 0.73392 and Recall of 0.74194 compared to the existing schemes.
Recent advances in Large Language Models (LLMs) have positively and efficiently transformed workflows in many domains. One such domain with significant potential for LLM integration is the Internet of Things (IoT), wh...
详细信息
Determining Worst-Case Execution time (WCET) is essential for temporal verification of real-time and embeddedsystems. These systems are designed to meet the stringent timing constraints imposed by the regulations. If...
详细信息
ISBN:
(数字)9781665497992
ISBN:
(纸本)9781665497992
Determining Worst-Case Execution time (WCET) is essential for temporal verification of real-time and embeddedsystems. These systems are designed to meet the stringent timing constraints imposed by the regulations. If a system gets delayed due to non-compliance with the deadline, it will lead to disastrous events. Worst-Case Data which gives maximum execution time, plays a vital role in the estimation of WCET. An evolutionary algorithm such as the Genetic Algorithm has been employed to generate the Worst-Case Data. The complexity of an evolutionary algorithm requires the use of several computational resources. This paper presents a novel method to replace the hardware and simulator used in the evolution process with machine learning models. This method reduces the overall time required to generate Worst-Case Data. Different machine learning models are trained to integrate with genetic algorithms. Our machine learning models are created using the Pygad Framework. The feasibility of the proposed approach is validated using benchmarks from different domains. The results show the speedup in the generation of Worst-Case Data.
According to the World Health Organization (WHO), 253 million individuals worldwide are visually impaired, including 36 million who are blind and 217 million who have moderate to severe vision impairment. The objectiv...
详细信息
Availing cloud computing services is the need of time. In contrast to being sold as a product, pooled resources, software, and data are made available to users as a network-based utility in cloud computing. real-world...
详细信息
Energy consumption is a significant concern for both small-scale embedded devices to large-scale data centers. The problem of minimizing the scheduling length of parallel applications with energy constraints in hetero...
详细信息
ISBN:
(纸本)9789819947546;9789819947553
Energy consumption is a significant concern for both small-scale embedded devices to large-scale data centers. The problem of minimizing the scheduling length of parallel applications with energy constraints in heterogeneous computingsystems has garnered widespread attention. The crux of this problem lies in the rational conversion of energy constraints of the application to the energy constraints of each task, utilizing dynamic voltage and frequency scaling (DVFS) technology. To determine the energy constraints of each task, the commonly used method is the task energy pre-allocation strategy. Previous studies only focused on energy allocation for individual tasks, neglecting the consideration of task levels. However, an equitable energy allocation for each task is not necessarily the optimal approach. If the levels of tasks in energy allocation are taken into account, more energy can be allocated to tasks that have a greater impact on the overall application scheduling time, thus effectively reducing the application scheduling time. Therefore, we have proposed a task level-aware scheduling algorithm that allocates energy to each level based on the proportion of the minimum energy consumption for the level, then, the algorithm assigns energy to each task based on the proportion of the task's weighted time in the current level. Extensive experimental results demonstrate that our approach allocates energy reasonably and achieves better performance, effectively reducing the application's schedule length.
To ensure high Quality of Service (QoS) for real-time IoT applications, it is essential to devise efficient status update strategies that improve the information timeliness at the destination (data consumer). Recently...
详细信息
ISBN:
(纸本)9798350378412
To ensure high Quality of Service (QoS) for real-time IoT applications, it is essential to devise efficient status update strategies that improve the information timeliness at the destination (data consumer). Recently, a variety of metrics, such as Age of Information (AoI) and its variations, have been developed to quantify the information timeliness, advancing the design of status update strategies in diverse IoT systems. In this paper, we propose a novel context-aware metric, the Value of Context-Aware Information (VoCAI), which evaluates information timeliness not only based on time latency but also considers the impact of the context (related to the content observed by the destination) on the status update requirements. Meanwhile, considering an IoT systems with constrained communication resources as an illustration, we formulate the VoCAI-oriented status update procedure as a Restless Multi-Armed Bandit (RMAB) problem and establish the indexability of the problem in a two-dimensional state space, guaranteeing the existence and asymptotic optimality of the Whittle index solution. To address the challenges arising from the unknown environmental dynamics, we further propose a Deep Q Network (DQN) algorithm to learn the Whittle index for the optimal scheduling policy. Finally, simulations are carried out to validate the effectiveness of our proposed status update strategy.
暂无评论