The issue of transcribed digit acknowledgment has for some time been an open issue in the field of example order. A few examinations have demonstrated that Neural Networks offer excellent performance in data classification. This project presents a reliable and efficient method for recognizing handwritten digits using Convolutional Neural Networks (CNN). CNNs outperform traditional neural networks due to their spatial feature extraction capabilities and improved computational efficiency. Using the MNIST dataset, which includes 70,000 grayscale images of digits (0–9), the CNN model is trained and tested to classify images with high accuracy. This is essentially a 10-class classification problem using deep learning frameworks like TensorFlow and Keras, with support from libraries such as NumPy and Pandas.
Monitoring and managing traffic is a critical need in urban areas where road congestion and violations are increasing. Traditional methods for vehicle counting, like manual surveys or sensor-based systems, are inefficient, costly, and prone to error. This project proposes a Real-Time Vehicle Detection and Counting System using OpenCV and computer vision techniques to automatically detect and count vehicles in traffic video feeds. The system leverages frame-by-frame analysis and object detection models to identify vehicles and increment counters as they cross a virtual detection line. It can be deployed for traffic management, road analytics, or surveillance in smart city environments.
Speech is one of the most natural forms of human communication, carrying not only linguistic information but also emotional states. Recognizing emotions from speech can enhance human-computer interaction, virtual assistants, healthcare, education, and security systems. Traditional machine learning methods often struggle to capture the complex temporal and spectral features of speech. In this project, a deep learning-based approach is used to detect emotions (such as happiness, sadness, anger, fear, neutral, etc.) from audio signals. The system preprocesses speech data into spectrograms or Mel-frequency cepstral coefficients (MFCCs), then applies Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN/LSTM), or hybrid architectures to classify emotions with higher accuracy.
The increasing global demand for renewable energy has made accurate prediction of solar energy generation crucial for efficient energy management and grid stability. This project focuses on developing a deep learning-based predictive model to forecast solar energy output based on historical and environmental data. By leveraging advancements in neural networks, the system aims to address the challenges posed by the intermittent and weather-dependent nature of solar energy. The proposed approach employs Recurrent Neural Networks (RNNs), particularly Long Short-Term Memory (LSTM) networks, to capture temporal dependencies in time-series data, such as solar irradiance, temperature, and weather conditions.
In India, six two-wheeler riders die every hour in road accidents. Also, we have seen that during this pandemic people wear masks, and to avoid congestion they do not wear helmets which attracted our concern and we decided to work on a project where these helmetless people can be penalized for violating traffic rules. To achieve an efficient helmet detection model, we have used the YOLOv5 object detection model using transfer learning. Further to check whether the biker is wearing a helmet or not we are using two methods, one being checking for overlapping between bounding boxes and the second method is, checking if a helmet exists in the specified range of coordinates above the motorcycle. Our model gives a mAP of 0.995 and to the best of our knowledge, we used overlapping methods for interlinking objects for finding the person not wearing a helmet. For number plate recognition we are using EasyOCR.
Social media platforms like Twitter, Facebook, and Instagram have become essential spaces for users to share opinions, feedback, and emotions. Monitoring these sentiments can provide valuable insights for businesses, governments, and researchers. This project aims to develop a Python-based Sentiment Analysis System that automatically analyzes textual posts from social media and classifies them as positive, negative, or neutral. The tool uses Natural Language Processing (NLP) and Machine Learning models to interpret the emotional tone of social media content in real time.
Considering a growing number of criminal acts, there is an urgent need to introduce computerized command systems in security forces. This study presents a novel deep learning model specifically developed for identifying seven different categories of weapons. The suggested model utilizes the VGGNet architecture and is implemented utilizing the Keras architecture, which is built on top of the TensorFlow framework. The model is trained to recognize several types of weapons, including assault rifles, bazookas, grenades, hunting rifles, knives, handguns, and revolvers. The training procedure involves creating layers, executing processes, saving training data, determining success rates, and testing the model. A customized dataset, consisting of seven different weapon categories, has been meticulously chosen and organized to support the training of the proposed model network. We do a comparative study using the newly created dataset, specifically comparing it with established models such as VGG-16, ResNet-50, and ResNet-101. The suggested model exhibits exceptional classification accuracy, obtaining a remarkable 98.40%, outperforming the VGG-16 model (89.75% accuracy), ResNet-50 model (93.70% accuracy), and ResNet-101 model (83.33% accuracy). This research provides a vital viewpoint on the effectiveness of the suggested deep learning model in dealing with the complex problem of weapon classification, presenting encouraging outcomes that could greatly improve the capabilities of security forces in countering criminal activities.
As per the previous year’s report concerning to road crashes indicates that the principal cause of such a fatal road accidents is because of negligence behavior as well as drowsiness of driver. This problem reveals the requirement of such a system that can recognize drowsiness state of driver and gives alert signal to the driver before the occurrence of any accidents. Therefore, this proposed work has established drowsy detection as well as accident avoidance system based on the eye blink duration. Here, first the open and close state of eye are detected based on the eye aspect ratio (EAR). Further, the blink duration or count during the changes of eye state from open to close are analyzed. Then, it identifies the state of drowsiness, when blink duration becomes more than a certain limits and sends the alert message to the driver through the alarm. Our developed system has shown the accuracy of 92.5 % approx on yawning dataset (YawDD).
Stress is a mental or emotional state brought on by demanding or unavoidable circumstances, also referred to as stressors. In order to prevent any unfavorable occurrences in life, it is crucial to understand human stress levels. Sleep disturbances are related to a number of physical, mental, and social problems. This study’s main objective is to investigate how human stress might be detected using machine learning algorithms based on sleep-related behaviors. The obtained dataset includes various sleep habits and stress levels. Six machine learning techniques, including Multilayer Perception (MLP), Random Forest, Support Vector Machine (SVM), Decision Trees, Naïve Bayes and Logistic Regression were utilized in the classification level after the data had been preprocessed in order to compare and obtain the most accurate results. Based on the experiment results, it can be concluded that the Naïve Bayes algorithm, when used to classify the data, can do so with 91.27% accuracy, high precision, recall, and f measure values, as well as the lowest mean absolute error (MAE) and root mean squared error rates (RMSE). We can estimate human stress levels using the study’s findings, and we can address pertinent problems as soon as possible.
This project proposes an automatic face naming system using discriminative affinity metrics learned from weakly labelled images. The goal is to accurately assign names to faces in a collection of images where the labels are incomplete or imprecise. By leveraging weakly labelled data, the system reduces the need for extensive manual annotation, making it scalable for large datasets. The approach involves constructing a discriminative affinity metric that models the relationship between face pairs based on visual features and available weak labels. A metric learning framework is employed to refine these affinities, enhancing the separability of different identities in the feature space. The methodology includes data preprocessing, feature extraction using deep learning models, and affinity metric learning through supervised and semi-supervised strategies. Experimental results demonstrate improved face naming accuracy compared to traditional methods, showcasing the effectiveness of the proposed technique in handling weakly labelled datasets. This system has potential applications in media analysis, security, and large-scale image retrieval systems.
Air and noise pollution are two of the most harmful environmental threats in urban areas, impacting both human health and ecosystem balance. This project presents an real-time monitoring system that measures air quality (e.g., PM2.5, CO2) and ambient noise levels. The system collects data through smart sensors and transmits it to a centralized server for visualization and analysis. The goal is to provide accurate, location-based pollution data that can assist governments, environmental agencies, and the public in understanding pollution trends and taking preventive actions.
