International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 03 | Mar 2025
p-ISSN: 2395-0072
www.irjet.net
AUTOMATIC BREAKING SYSTEM USING ARDUINO Nagendrababu N C1, Aishwarya Y S2, Lakshmi P N3, Mohammad Umar Farooq4, Vilas S5 1Assistant Professor, Dept of CSE(AI&ML), SJCT Chikaballapur INDIA, 2,3,4,5 Dept of CSE(AI&ML), SJCT Chikaballapur INDIA.
-------------------------------------------------------------------------***-----------------------------------------------------------------------ABSTRACT: With the increasing occurrence of road accidents, the necessity for integrating automated braking systems within vehicles has become imperative. This study presents a sophisticated deep learning framework specifically designed for the classification of various driving scenarios. The suggested model is founded on the VGGNet architecture and is executed using Keras, which functions on the TensorFlow platform. It has been developed to detect situations such as sudden obstacles, pedestrian crossings, vehicle proximity, wet roads, sharp turns, traffic signals, and lane departures. The training process encompasses constructing layers, executing operations, preserving training data, evaluating performance metrics, and validating the model. A specialized dataset, systematically curated to cover all seven driving scenarios, has been utilized to enhance the model’s learning capabilities. A comparative analysis is conducted using this dataset, benchmarking the proposed model against established architectures such as VGG-16, ResNet-50, and ResNet-101. The findings demonstrate that the proposed model attains an outstanding detection accuracy of 98.40%, surpassing VGG-16 (89.75%), ResNet-50 (93.70%), and ResNet-101 (83.33%). This study highlights the effectiveness of the developed deep learning approach in addressing the intricate task of driving scenario classification, presenting promising results that could significantly strengthen vehicles' ability to prevent accidents.
thereby facilitating prompt intervention. These systems can help reduce criminal activities by enabling security personnel to address incidents in real-time, rather than responding after they have already occurred. This study supports the creation of an intelligent system that employs sophisticated software to quickly notify security personnel when hazardous objects are detected, thus improving crime prevention strategies. Deep learning has gained significant acclaim for its capacity to improve security and surveillance operations. This particular domain of machine learning utilizes numerous layers of non-linear processing units to extract and enhance features. It emphasizes representation learning by examining various levels of data attributes, rendering it especially effective in image and video processing. In security applications, models based on deep learning can proficiently analyze surveillance footage to detect potential threats. The process of feature extraction in image processing entails calculating pixel density metrics and identifying unique patterns such as edges, textures, and shapes. Among the most prevalent architectures in deep learning for image classification is the Convolutional Neural Network (CNN). CNNs are composed of several layers, including convolutional, pooling, activation, dropout, fully connected, and classification layers, all of which play a role in learning hierarchical features from unprocessed images. Owing to their remarkable accuracy and efficiency, CNN-based models have emerged as a favored option for object detection and recognition tasks, including the identification of firearms in surveillance videos. In modern society, the prevalence of criminal activities frequently associated with portable firearms necessitates that law enforcement agencies implement sophisticated monitoring systems. Research has consistently demonstrated the significant role of handheld weapons in various unlawful activities, including theft, unauthorized hunting, violent assaults, and acts of terrorism. One potential solution to mitigate such crimes is the deployment of intelligent surveillance mechanisms capable of early threat detection, thereby allowing security forces to take immediate action before an incident escalates. However, identifying weapons in real-time presents unique challenges, such as occlusions, object similarities, and background complexities. Occlusion occurs when a portion of the weapon is obscured by objects or body parts, making detection difficult. Object resemblance issues arise when everyday objects, such as mobile phones, tools, or clothing, visually
Key Words: Keras, Tensorflow, VGGNet architecture, Constructing Layers, Performance Matrix.
INTRODUCTION With advancements in science and technology, surveillance cameras have become an integral tool in crime prevention. Security personnel hold the responsibility for ensuring safety and protection carefully monitoring and deploying camera networks across multiple locations. Conventionally, incident analysis involves security teams reaching the crime scene, reviewing recorded footage, and gathering relevant evidence. This process, however, is reactive and often leads to delays in responding to potential threats.There has consequently been a growing focus on the importance of proactive surveillance systems capable of identifying potential threats in real-time,
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 100