Bo Mei, College of Science and Engineering, Texas Christian University, Fort Worth, TX 76129, USA
The classification of videos has become increasingly important in the field of data science research, as it has numerous practical applications in modern society. Compared to image classification, video classification poses a significantly greater challenge. One of the most obvious difficulties is that video classification tasks require more powerful computers due to the large number of features that need to be computed. Additionally, conventional 2D Convolutional Neural Networks (2D CNNs) are not effective in handling such tasks. This paper proposes a novel 2-layer Convolutional Neural Network (CNN) architecture for action recognition that addresses these challenges. The proposed architecture achieved a high test accuracy of 79.66% for classifying large video clips. The results indicate the effectiveness of the proposed approach for video classification tasks.
Neural networks, video classification, action recognition.
Dr B.V. Rama Krishna1, B. Sushma2 and V. ChandraSekharRao3, 1,3Department of Computer Engineering, Aditya College of Engineering, Surampalem, India, 2IT-Department, MLRIT, Hyderabad, India
The smart devices communication and data sharing becoming vital part of today’s digital life. The interaction between electronic devices and computational devices made effectively easy with IoT technology. In Metro cities managing the traffic and avoiding traffic congestions becoming a serious issue for police department. In this paper RFID based communication among IoT devices of Traffic Management System and its governance overviewed. Automation of major traffic management system highlighted with RFID technology perspective. The economic benefits and limitations of RFID application over Traffic Systems are coined in this paper. The data mining technologies empowering the data analysis over RFID systems improved the automation process of traffic management systems. The decision support system is boosted with data mining interface with RFID technology in traffic systems.
RFID, IoT, Sensors, Transmitters, Receivers, Data Mining Techniques.
Diego Vallarino, Independent Researcher, Madrid, Spain
Businesses generate massive volumes of data that is frequently underused due to the fast rise of the Internet and information technology. Translating raw data into information and knowledge that drives decisionmaking unlocks its worth. Machine Learning (ML) algorithms can analyze big datasets and produce accurate predictions. ML has been used in market segmentation, customer lifetime value, and marketing strategies. This article reviews marketing ML methods such Support Vector Machines, Genetic Algorithms, Deep Learning, and K-Means. ML is used to consumer behaviour analysis, recommendation systems, andbankruptcy prediction. Kernel SVM, DeepSurv, Survival Random Forest, and MTLR survival models are compared to predict bank failures. The data show that consumer purchase history, promotions, discounts, pleasant customer experiences, age, online activity, and hobbies impact purchase time. These insights help marketers improve conversion rates.The DeepSurv model predicts purchase completion best in the research.
Survival Analysis, Machine Learning for survival, marketing analytics.
Shrey Yagnik1, Mayank Kumar2, Priyanka Singh1 and Manjunath Joshi1, 1Dhirubhai Ambani Institute of Information and Communication Technology, Gandhinagar, Gujarat, India, 2Indian Institute of Technology Jammu, Jammu, India
Deep learning models have been used in numerous security-critical settings since they have performed well on various tasks. Here, we study a kind of attack on Federated Learning (FL). FL has become a popular distributed training method because it enables users to work with large datasets without sharing them. Once the model has been trained using data on local devices, only the updated model parameters are sent to the central server. The FL approach is distributed. Thus, someone could launch an attack to influence the models behavior. In this paper, we conducted the study for a Backdoor attack where we added a few poisonous instances to check the models behavior during test time. Here, the poisoning could pertain to a single class or multiple classes. We conducted various experiments using the standard CIFAR10 dataset to alter the models behavior. We found out that the expected behavior of the model could be compromised without having much dif erence in the training accuracy.
Federated learning, Backdoor Attacks, Backdoor Trigger.
Yao Cheng1, Junghee Kim2, 1Arnold O.Beckman High School, 3588 Bryan Ave, Irvine, CA 92602, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
This research focuses on the use of AI-powered badminton video analysis to enhance gameplay analysis andtraining. The technology utilizes artificial intelligence and machine learning algorithms to analyze varJaccardSimilarity Index aspects of badminton game footage, including player movements, shot selection, and game strategy. It provides personalized feedback and recommendations for improvement to players and helps coaches identifypatterns and trends in their players performance. The use of lightweight models, such as YoloV5, is essential for real-time video analysis due to the need for high- speed processing. These models, including MobileNets, Ef icientNets, SqueezeNet, Shuf leNet, and Tiny-YOLO, aredesigned to be lightweight and optimized for speed while maintaining high accuracy. The Yolo model, a one-shot learning model, is particularly suitable for real-time object detection tasks due toitsimpressive speed and accuracy. It uses anchor boxes and data augmentation techniques to quickly learnandrecognize objects with a high level of accuracy. The research process involved the collection of badminton match videos, preparation of datasets, and testing of theYoloV5 models performance. Precision and Jaccard Similarity Index metrics using Jaccard similarity were usedtoevaluate the models performance in detecting player positions and court boundaries. The results showed that the YoloV5 model performed better when tested on the video data with less clear imageof judges in the background. The Jaccard Similarity Index metrics using Jaccard similarity demonstrated improvedperformance when tested on generalized video data was used to accurately evaluate the overlap between predictedand ground truth Overall, AI-powered badminton video analysis has the potential to revolutionize the way badminton is playedandcoached. The use of lightweight models like YoloV5 enables faster and more ef icient real-time analysis, makingit practical for a wide range of applications in badminton.
Artificial Intelligence, Badminton, Video Analysis, Machine Learning.
Kaige Bao1, Ang Li2, 1Basis International Hangzhou, No.9 Yulin Road, Shangcheng District, Hangzhou, Zhejiang, China, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
This paper investigates the efficacy of machine learning algorithms for the detection of Distributed Denial of Service (DDoS) attacks . The study explores different approaches, including Support Vector Machines (SVM), logistic regression, and decision trees, and evaluates their performance using metrics such as accuracy, precision, recall, and F1-score . The results demonstrate the effectiveness of SVM models with polynomial or radial basis function (RBF) kernels, logistic regression models with a polynomial degree of 4, and decision tree models with depths exceeding 10 . These algorithm configurations exhibit promising potential in mitigating DDoS attacks and safeguarding network infrastructures . However, limitations such as dataset availability, imbalanced data, and the focus on offline detection warrant further research. Enhancements in these areas can lead to more robust and efficient DDoS detection systems. The findings of this study contribute to the advancement of network security and offer insights for organizations aiming to counter the growing threat of DDoS attacks.
Machine Learning, DDoS Attacking, Algorithms, Computer Science.
Ayoub EL IDRISSI, Abdelfatteh Haidine, National School of Applied Sciences, University Chouaib Doukkali, El Jadida, Morocco
This article explores the real-world benefits of intelligent traffic in seaport operations and the impact of integrating IoT and automation using deep learning models. By leveraging advanced monitoring, control, and anagement systems, seaports can optimize traffic flows, improve efficiency, and reduce congestion and waiting times. The study evaluates the impact of variables such as IoT usage, level of automation, median time spent in port, ship and port characteristics, and more on port performance. The deep learning model trained can be used to analyze historical and real-time data, identify patterns, and predict traffic flows accurately. The research findings highlight the variables that have the greatest impact on port performance and discuss how managing these variables effectively can optimize port processes and reduce vessel turnaround times. The study also emphasizes the role of IoT usage, automation, and data analytics in making informed decisions for optimizing port operations. The results demonstrate the tangible benefits of implementing intelligent traffic and automation in seaport operations, offering valuable insights to decision-makers and port managers to enhance operational efficiency and deliver higher quality services. Embracing intelligent traffic and leveraging IoT and automation can unlock opportunities to improve competitiveness, sustainability, and overall operational performance in the dynamic maritime industry.
Intelligent Traffic, IOT, Automation, Vessel, Seaport, Performance, Time Management, Neural Network.
Daniel Miao1, Tyler Liu2, Andrew Park3, 1The Harker School, 500 Saratoga Ave., San Jose, CA 95129, 2The Webb Schools, W. Baseline Road, Claremont, CA 91711, 3Computer Science Department, California State Polytechnic University, Pomona, CA91768
The advent and spread of the internet has caused many users to favor the convenience and breadth of reportingthat online news of ers, whether it be from media companies or social platforms, which in turn has led tothemonetization and corruption of said stories . Large, company-owned news sites each try to appeal to only afewgroups across the political spectrum, oftentimes sacrificing the indif erence and integrity which serve as the tenetsof honest journalism. We propose to aid in solving this problem a Chrome extension which serves to provide metrics, information, and line-by-line analysis of article text in order to help readers stay aware and healthily skeptical . Using machine learning (ML) as well as traditional algorithms, we aim to provide key info on the article’struthfulness as well as the source’s bias and ownership . In this project, we used 3 main models, each to detect fake news, political leaning and sentiment, in addition to traditional criteria such as readability, # of words, andtime to read. All of our models performed well both theoretically and practically, giving above 80%accuracy on all occasions.
Artificial Intelligence, Fake News, Chrome Extension, Text Analysis
PLTA Gunerathne, General Sir John Kotelawala Defense University, Kandawala Road, Ratmalana, Sri Lanka
Cities rising urbanization and population expansion have resulted in an increase in trash creation, offering considerable difficulties to traditional waste management strategies. To address these environmental problems and promote sustainability, this project proposes the creation of a Smart Waste Management System that makes use of the Internet of Things (IoT) and Machine Learnings synergistic potential. The system intends to optimize garbage collection, decrease environmental impact, and build a more sustainable urban environment by deploying IoT-enabled smart waste bins, real-time data collecting, and predictive analytics. The Smart trash Management System intends to improve trash management techniques in big communities via the merging of data-driven decision-making and community interaction.
Urbanization, Population Expansion, Waste Management Strategies, Sustainability, Predictive Analytics, Real Time Data Collection
Aaron Li1, Kian Azadi2, 1Francis Parker School, 6501 Linda Vista Rd, San Diego, CA 92111, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
Drunk driving is a public health crisis not only in the United States but worldwide that is actively being addressed, with many dif iculties along the way. Sober guardian intends to help address this with the usage of MachineLearning and Computer Vision to quickly detect if the driver is impaired and stop them before they start driving. Through the usage of facial recognition, machine learning, and in the future chemical sensors, it is believed tobeef ective and ef icient to detect the drivers level of impairment. There were many challenges encountered, suchasimage unclarity, small amounts of data, and usability. Experiments were run to see how the model would react totwo dif erent forms of input: images and videos. For both experiments, it was found to predict sobriety with 86%- 87.5% accuracy. The video experiment was unable to be tested for people under the influence due to lack of suf icient datasets. Sober guardian, while in the early stages of development, has shown to be an ef ective solutiontodetect the driver’s impairment and stop them from driving under the influence.
Machine Learning, Drunk Driving, Internet-Of-Things, Facial Recognition
Cheng Xiang and Zheng Yangfei, Department 8 of System, North China Institute of Computing Technology
With the development and application of Internet technology, a large amount of text information has become an important source of information for people. However, the diversity of information sources and the complexity of expression make it one of the hotspots in current natural language processing research to efficiently and accurately extract valuable information from a large amount of text. In view of this problem, this paper takes the extraction of stock increase or decrease information as an example to study how to automatically extract unstructured information from the announcements of listed companies to help investors better obtain information and make wise investment decisions. This paper proposes a rule-based unstructured information extraction model that combines web scraping technology and NLP technology to extract features from HTML documents, automatically identifying and extracting information related to stock increase or decrease. The model has strong practicality and promotion value and has broad application prospects in the field of investors and related research. At the same time, this paper also recognizes that the model still has some limitations and shortcomings, which need to be further explored and improved.
Unstructured Information, NLP, Automatic Identification, Feature Extraction
Mutiu Iyanda Lasisi1 and Kingsley Mawuli Kesseh2, 1Department of Media Communications, HSE National Research University, Moscow, Russian Federation and 2Department of World Economy and International Affairs, HSE National ResearchUniversity, Moscow, Russian Federation
This paper investigates the positions of state and non-state actors during the ban of Twitter by the Nigerian government between June 4, 2021 and January 13, 2022. It answers questions about whether non-state actors resisted the ban due to Twitters contributions to democratic, civic space, and economic growth, and whether state actors considered the medium a threat to the countrys unity and sovereignty with supportive evidence. Conflict theory and the Advocacy Coalition Framework are used as theoretical and analytical frameworks to explore the different beliefs and interests of the actors involved. The study found that the ban generated a series of reactions from these actors, leading to the development of three forms of narratives from the policy and advocacy coalition beliefs. The study concludes that examining the views of state and non-state actors during the ban period provides insights into the different interests, beliefs, and values that shape policy conflicts.
Conflict, Democracy, Twitter, Nigeria, Policy Actors
Jialiang Liu1, Moddwyn Andaya2, 1La Salle College Prep, 880 East Sierra Madre Boulevard, Pasadena, CA91107, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
Currently, there is a significant rise in the popularity of 3D donation programs, wherein numerous participantsactively engage in mutual acts of charitable giving. Drawing upon extensive research conducted within these virtual philanthropic communities, we have developed and implemented a visualization system aimed at providing userswith an immersive and playful experience while exploring and navigating large-scale 3D donation networks . Our design leverages familiar three-dimensional representations to introduce novel techniques for comprehendingthe interconnectedness of complex donation structures, supporting visual analysis and search capabilities, as well as automatically identifying and visualizing philanthropic clusters . Through public installations and controlledstudies, our system has demonstrated its usability, ability to facilitate discovery, and potential for fosteringenjoyable and socially engaging philanthropic activities.
3D Modeling, Donation, Computer Science.
Yuanxin Jia1, John Morris1, 1University High School, 4771 Campus Dr, Irvine, CA 92612, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
The process of Scouts within the Boy Scouts of America ranking up to First Class, and eventually Eagle, canbequite gruesome and challenging . Though there are various resources online ranging from articles and videosthat explain how to potentially speed up the ranking process, none of them actually give an ef ective solution totheproblem. Scouts nowadays lack things like attention span because of the harmful side of social media, especiallyTikTok . They don’t want tips and tricks on how to rank faster, and they also don’t want to flip through massivetextbooks to find the answers to requirements. They just want straightforward and quick answers to their problemswithout any stalling. This is exactly what the Trail to First Class app is. It gives them quick answers to theirproblems without any ties. Though, the app is not perfect. There is a severe limitation that if some requirementsrequire scouts to, for example, identify things around their geological area, the app cannot help thembecauseproviding the answers to every single possible geological area would be quite inef icient. Though something like thismight be possible in the future, it definitely cannot come now. Despite the limitations, the app is more than enoughto provide scouts with all the resources they need in order to accelerate their ranking process.
Boy scouts, Trail to first class, Unity engine, Eagle.
William Liang1, Jonathan Sahagun2, 1Portola High School, 1001 Cadence, Irvine, CA 92618, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
This project focuses on the development of an af ordable and accessible solution for early wildfire detection . Byleveraging a network of sensors spread across a wide area, the project aims to empower individuals andcommunities with the ability to proactively detect fire sources . Advanced algorithms analyze environmental factors such as temperature, humidity, and wind speed to enhance the accuracy and timeliness of wildfire detection. By addressing the limitations of existing methodologies, this project contributes to improving wildfireprevention and mitigation ef orts . Through cost-ef ective technology and widespread implementation, it aims tocreate safer and more resilient communities in the face of the increasing threat of wildfires .
Wildfire Detection, Sensor Network, Environmental monitoring, Algorithms.
Zeyu Zhang1, John Morris2, 1Santa Margarita Catholic High School, 22062 Antonio Pkwy, Rancho Santa Margarita, CA92688, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
The goal of this project is to bring up awareness on the safety of endangered species by creating an educational game called "Panda." . By selecting a panda as the main character, which symbolizes endangered species, thegame aims to engage players in learning about the challenges these animals face . In the game, playersmanipulate a panda and strive to survive as long as possible while encountering various threats such as caraccidents, predators, and hunters. To enhance the educational aspect, a fact sheet pops up after each panda death, informing the player about thehuman causes behind the pandas demise and suggesting preventive measures. For instance, if a panda dies in acaraccident, the fact sheet would highlight the staggering statistic of 1 to 2 million wildlife deaths caused by caraccidents annually in the U.S. It would also recommend supporting the Roadless Rule as a means to reduce suchaccidents . Although the proposed solution is promising, its ef ectiveness still requires validation as the game is relativelyunknown and has been played by only a few individuals. Future plans involve conducting surveys among a largerplayer base to assess the games impact and ascertain whether the desired awareness and behavioral changesregarding endangered species conservation are achieved.
Environment, Wildlife Safety, Endangered species, Educational Game, Survival Quest
Tahirou DJARA, Laboratoire d’Electrotechnique de Télécommunication et d’Informatique Appliquée and Institut d’Innovation Technologique
In this paper, we have developed a new measure called (Radio Online Audience) of the audience rate of a web radio station. The is a method that uses real-time audience tracking and measurement technologies to provide a more accurate estimate of the audience rate of web radios. It is a composite formula that incorporates the radio broadcast frequency, the number of radio listeners and the average listening time of all radio listeners. In order to implement the new measure, an experiment is being conducted on a family of twelve radios from the “AfricaWebRadio” ecosystem. We have also extended the experiment on the results of the ranking of radio stations (March 2023) carried out by the ACPM (Alliance for Press and Media Figures). The experimental results are promising.
Ethan Chen1, Jonathan Sahagun2, 1Sage Hill School, 20402 Newport Coast Drive, Newport Coast, CA 92657, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
Many people today lack basic first aid knowledge and emergency preparedness . Official first aid training requires too large of a time commitment and lacks concision, causing many learners and lifeguards-in-training to be disincentivized. In order to combat this problem, we created a first aid simulative video game in which the player acts as a lifeguard and needs to carry out various procedures such as CPR and water rescue . We used Unity Game Engine, Visual Studio 20, C# programming, mixamo.com, and the Unity Asset Store to construct our game . A key challenge we underwent was animating the various models due to the complex nature of creating animations. Instead of making the animations ourselves, we used mixamo.com to download already-published animations to import into our own game. In order to enhance the user experience, we experimented with the camera speed and hitboxes in order to get all the details right. Ultimately, Lifeguard Simulator provides a concise, captivating, and enjoyable first aid learning experience that is not available elsewhere.
Unity Game Engine, Lifeguard Training, 3D Interactive Simulation, Animation Rigging.
Chunxuan Zhang1, Andrew Park2, 1Arnold O’ Beckman High School, 3588 Bryan Ave, Irvine, CA 92602, 2Computer Science Department, California State Polytechnic University, Pomona, CA91768
There has not been a sustainable amount of leaders matching up to the demand for such skill by 2030. Leader’sLegacy has been created to help alleviate the issue by presenting educational questions on leadership principleswithin an interactive game . Our solution is built with a combination of quiz-style questions and scenarios that are tied to player progression for the combat system . We also integrated a cloud-based backend via Firebasetoallow users to access their information from anywhere for greater accessibility.
Leadership, Unity, Firebase, Education.
Kingsley Ofosu-Ampong, Department of Information Technology Studies, UPSA, Ghana
The public sector worldwide is undergoing restructuring and optimization efforts to improve efficiency and cost-effectiveness, with a particular focus on procurement. Prior studies have highlighted the importance of the right sourcing decisions and the negative consequences of poor decisions. However, sourcing decision in the era of digital transformation has received less attention in effective sourcing decisions. The purpose of the study is to assess the ongoing digital transformation effect of sourcing decisions on the performance of procurement functions. The objective is to analyse the criteria employed in selecting suppliers to make a call for digital transformation initiatives in Ghana’s procurement processes. The study found that factors such as after-sales services, financial capacity, and compliance with rules and regulations had a strong and statistically significant relationship with procurement performance. It also revealed that supplier evaluation and selection influenced the organisation’s operations by delivering value for money, transparency, reducing corruption-related costs, and satisfying top management and stakeholders. From the findings, it is worth noting that in todays business landscape, digital transformation can greatly influence the sourcing and procurement process.
Procurement Process, Digital Transformation, Sourcing Decision, Supply Chain.
Prof. Anyu Lee, Medical device regulatory research institute ,Sichuan University
Digital security risk is an emerging risk source of medical devices with adoption of various digital systems such as automation, data processing and artificial intelligence. The digital security risk will seriously undermine the device safety and may lead to the life threating consequences. This paper presents an effective method of assessment of digital security risk based on the real-time core dump. This method is both low cost to implement and high efficiency to detect. This method uses a debugger like tool to analyze the core-dump information of the digital system implemented on the medical devices and look for the suspicious mal-functional code sections. Those code sections will then be simulated with the testing system provided by the manufacturers of the medical devices to identify the risks. This method can be effectively in most client-service application and cloud applications.
Medical devices, digital security, digital risk, digital safety, real-time digital record.
Abdulrahman S Almuhaidib, Sunday Olusanya Olatunji, and Abdulmajed A. Aldulaijan, College of Computer Science and Information Technology, Imam Abdulrahman Bin Faisal University, P.O. Box 1982, Dammam 31441, Saudi Arabia
The physical and online market is filled with products that are claimed to be genuine. The organizations that own the rights to these products paid huge amounts of money for these products to be delivered. Because of their cheap prices and their look very similar to the original products, counterfeits are invading the market. Unfortunately, some sellers are exploiting this by marketing their fake items claiming that they are genuine, unaware or sometimes not caring about the consequences. Therefore, this paper will study the importance of implementing a public supply chain ledger that is easy to access for everyone to investigate if the product in front of them is as genuine as it is claimed to be. For that, Blockchain technology is proposed here since it is built to be available to all authorized parties, easy to access and immutable.
Blockchain; digital ledgers; supply chain management; counterfeit products; Hyperledger Sawtooth.
Mousa Jari1, 3, Kovila Coopamootoo2 and Rasha Ibrahim1, 1School of Computing, Newcastle University, Newcastle upon Tyne, UK, 2Department of Informatics, King’s College London, London, UK, 3College of Applied Computer Science, King Saud University, Riyadh, Saudi Arabia
Amid growing concerns about security and privacy, and their impact on decision-making, researchers have sought to understand the reasons behind users’ seemingly risky behaviour in disregarding security advice. In this study, we delve into the perceptions of security experts on end users’ threat models and their cybersecurity practices and habits. This research explores the perceptions of security and privacy experts regarding end users’ threat models and their behaviours in relation to cybersecurity. A survey was conducted with 55 experts, including 27 females and 28 males, to gain insights into end users’ habits, practices, and feelings from the perspective of security experts. The study reveals that end-users express moderate concern about privacy and security while carrying out their daily tasks. However, security experts believe that end-users tend to be passive towards organisational security policies, and their lack of knowledge about these policies which may lead to negative feelings. Additionally, experts perceive that end-users may be unaware of security measures, have difficulties understanding security concepts, and are at high risk of falling victim to phishing attacks by opening attachments and clicking on unknown links.
Security, Privacy, Policies, Phishing, Experts & End-users.
Junyu Qian1, Jonathan Sahagun1, 1Portola High school, 1001 Cadence, Irvine, CA 92618, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
I made a device that will make the phone that is connected to the device through Bluetooth play a note when the device detects a movement. When the device was being shaken by someone, the accelerometer will detect an acceleration and it will make the phone that is connected with the device with Bluetooth play notes in the app. This paper discusses the development and application of a Bluetooth-enabled accelerometer-based movement detection system for musical interactions with mobile devices . The system consists of a device attached to an object and connected to a mobile device via Bluetooth . When the object is moved, the accelerometer detects the movement and triggers the mobile device to play a musical note through a dedicated app. The system was tested using different objects and movements, and the results showed that it was accurate and consistent. The paper highlights the potential uses of the system in music performance, education, and therapy, such as creating a new type of instrument, teaching musical concepts in an interactive way, and helping individuals with physical disabilities to engage in music making. Overall, the paper demonstrates the potential of Bluetooth-enabled accelerometer-based movement detection for musical interactions with mobile devices.
Fidget toys, Bluetooth-enabled accelerometer, Musical interaction.
Richard Feng1, Jonathan Sahagun2, 1,Troy High School, 2200 Dorothy Ln, Fullerton, CA 92831, 2Computer Science Department, California State Polytechnic University, Pomona, CA 91768
Distracted driving is an important problem that plagues the world, and is the cause of thousands of deaths in the United States. Distracted driving has cost an estimate of $44.2 billion to the US in damages. A way I can help solve this problem is creating a device that can detect if the driver is distracted. This device contains an Arducam Day n Night 1080p camera, a Raspberry Pi 3, and a buzzer. The device utilizes python and Mediapipe, a library which can locate important landmarks on the driver’s face . During development, I had to face many challenges such as the driver possibly looking down subtlety with their eyes, or the camera being positioned in a weird way. During experimentation, I tested the delay, and accuracy of the device. After 35 trials, the device is accurate around 94%, and the delay is around 9 seconds. Overall, this device can be used by everyone who wants to be kept responsible and alert while driving.
MediaPipe, Raspberry Pi, AI, Driving.
Copyright © VLSIA 2023