Crowd Real Time Video Classification, Count and Flow

• Main Author: Wee, Joel Hong Shen
• Format: Final Year Project
• Language: English
• Published: Universiti Teknologi PETRONAS 2020
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://utpedia.utp.edu.my/23053/1/FYP2%20Dissertation.pdf; http://utpedia.utp.edu.my/23053/

The need for smart surveillance systems is ever growing in the present days, involved in purposes such as security and marketing to track the movements of different classes of people. Our project in computer vision with deep learning is focussed on segregating the gender composition of people, while recognising and counting their flow of direction. The project will be used with reference to real-time video processing. The challenges/problem statement for the project is the lack of definitive methods to determine the direction of individuals, computationally expensive object detection models and lack of practical gender detection datasets. In this paper, the method of object detection with object tracking running in parallel is suggested to improve processing time of video frames, with a usage of OpenCV to identify existing, new and out-of-frame objects. A practical dataset of genders from crowd view to be used to fine-tune a pretrained object detection model is suggested for application as well.