Speakers
Description
Video surveillance plays a frontline role in security monitoring. The traditional human operation, that is time-consuming and prone to fatigue, is replaced by artificially intelligent modules that are powered by algorithms which enable classification of objects along with their spatial locations. However, a modular and contextual open-source solution to detect and track humans remains elusive. In this talk, we propose to demonstrate a cost-effective and efficient video surveillance system using Raspberry Pi. We begin with a neural network that solves a regressing problem and detects humans in a video stream that is received via RTSP protocol. To track the detected humans, we deploy a customized Kalman filter that can generate tracklets for the occluded humans in the video scene as well. Finally, a network that expands an image classification architecture along multiple axes – in space, time, width and depth – identifies activities of detected humans from a predefined pool of categories. This project can be further extended as an edge-device to include functionalities like anomaly detection, event notification, and integration with existing security systems.
Session author's bio
Priyam Chakraborty:
Priyam Chakraborty is an Aerospace Engineer with B. Tech., M. Tech. and PhD from Indian Institute of Technology Kharagpur. He has committed to the use of computational tools as a lead data scientist in a startup ecosystem, followed by post-doctoral fellowship at the University of Waterloo. He is currently working on smart active matter in the Indian Institute of Science. His fascination with collective intelligence, initially sparked by the efficiency of bird flocks, has turned into a belief in the power of biomimetic design that invokes automation and machine learning to unlock hidden patterns within complex datasets. By exploring, analyzing, and potentially challenging existing assumptions, Priyam aims to create affordable intelligent systems that can automatically analyze video data, extract meaningful insights about human activities, and potentially be applied in areas like security monitoring, human-computer interaction, and video surveillance.
Wajoud Noorani:
Wajoud Noorani is a committed data scientist at Changejar, where he operates as a full-stack developer. His responsibilities span the entire data lifecycle, from extraction and processing to manipulation and pattern discovery. With a keen interest in computer vision and natural language processing, he leverages his expertise to derive actionable insights from complex datasets. His work involves employing advanced machine learning techniques to solve real-world problems and drive innovation within the company. Wajoud is passionate about exploring the frontiers of AI, particularly in how it can enhance data-driven and economical decision-making and improve various business processes.
Any other info we should know?
The proposed talk directly aligns with the conference’s focus on open-source computing and benefits attendees in several key ways. Firstly, the project prioritizes open-source tools like Python libraries and pre-trained models, allowing for easy replication, customization and further development by the community. Secondly, the cost-effective and widely available Raspberry Pi platform makes the solution accessible to those with budget constraints. Furthermore, the demonstration delves into the practical implementation of integrating the minimal components of the latest algorithms within the Raspberry Pi environment. This provides a valuable roadmap and insights for attendees looking to implement similar functionalities in their own projects. By openly sharing design choices, challenges, and potential optimizations, this work fosters knowledge sharing and encourages further innovation within the open-source computer vision community. Ultimately, this project not only presents a functional video surveillance system but also serves as a valuable learning resource for attendees interested in leveraging open-source tools for real-world computer vision applications on resource-limited hardware platforms.
Special accommodations
Projector for presentation
Please confirm that there are included headshots of all speakers in their profiles | Yes |
---|---|
Agree to Privacy Policy and Notice | I agree |
Level of Difficulty | Intermediate |
In Person Attendance | In-person |
Social Media | linkedin.com/in/priyam-chakraborty-10412a86 |