Identification of Illicit Activities & Scream Detection using Computer Vision & Deep Learning
Rohan Mathur, Tejas Chintala, D Rajeswari
Abstract
Self-reliant technologies as well as fully automated technologies are the few technology behemoths that propel entire disciplines forward by various bounds. Significant advances in technology can be tagged to activities which are better handled and are more visible by means of automation. Machine learning is currently being used to make these computer algorithms usable to be smart enough to carry out tasks with right decision making, causing a massive reduce in human interference within redundant processes. Our proposal adheres to the following ideals: creating a solution that reduces manual work (physically and emotionally) for jobs that can be smoothly automated and processed and at the same time addressing the core statement at hand. At times like this, surveillance cameras serve an important role in ensuring people’s safety, but they are only multimedia entities with no intelligence. Because of the increased volume of data generated by security footage, automated video streaming have become necessary for automatically recognising abnormal occurrences. Along with this, audio data of these surveillances are hardly put to use. The project’s main goal is to promote safety by techniques incorporating deep learning to optimise the recording and evaluating crimes from sensory Closed-Circuit Television (CCTV), delegating the liability of recognizing illegal conduct to something like a workflow that can recognize trends and classify them for scanning. Along with this, it aims to detect screams from input audios in cases of failure that can correspond to crimes like rape. Scream Detection Model’s accuracy comes up till 0.95 in differentiating three categories.