People counting performance improvement using map-reduce architecture approach / by Afnan Hassan Saleh ; under the supervision of Prof. Amr Wassal, Prof. Elsayed Hemaye.

Thesis (M.Sc.)-Cairo University, 2022.

Bibliography: pages 94-98.

In this era, the concept of smart cities massively spreads to make human life easier and safer, as everything is monitored by surveillance cameras. Many successful business models are based on the idea of analysis of this huge amount of feeds from surveillance cameras. The evaluation parametric of computer vision applications changes to support the massive data. Nowadays, Computer Vision algorithms should be scalable, efficient, accurate, and low cost. Supporting the new evaluation parametric become a hot research field. Porting computer vision applications to Big-Data frameworks is one of the solutions to increase the scalability and throughput of these applications.
On the other side, the researchers and scientists recently had to use Deep Learning Algorithms which are simple and accurate, but it uses expensive GPU hardware to run with real time efficiency.
This research presents a method of parallelization of the People Count CV algorithm using Big-data processing frameworks, which randomly distribute the input data across the available processing units to utilize the efficiency of the throughput.
People Count is one of the computer vision algorithms, which hasn’t been marked as a solved problem yet, due to the challenges that exist in its internal components.
People Count algorithm consists of three main computer vision algorithms blocks, which are Pedestrian Detection, Multiple Objects Tracking and Counting algorithm. It follows the framework of Computer Vision Object Detection, Tracking with Counting algorithm as a post-processing part.
Our goal in this research is to parallelize the Pedestrian Detection block, which is huge in time complexity to enhance the efficiency of the Detection Tracking computer vision framework through The Big-Data framework.

في هذه الأطروحة عرضنا منهجيه جديده لزياده اداء خوارزميه عد الناس باستخدام نهج تقليل-خريطه البيانات الضخمه. الطريقه المعرفه في الاطروحه عباره عن تشغيل مجموعه عمليات اكتشاف بالتوازي وترسل النتيجه لطابور لتنفيذ عمليه تتبع. عند تشغيل اكثر من عمليه اكتشاف متوازيه الطريقه الجديده حققت زياده في السرعه و تقليل في كفاءه العد مقارنتا بالتشغيل الممتالي. علي سبيل المثال عند تشغيل عمليتين اكتشاف متوازيتين الطريقه الجديده حققت اربعون في المئه زياده متوسطه في السرعه و قللت كفاءه العد بنسبه سبعه و عشرون من مئه في المئه نسه متوسطه

Issues also as CD.

Text in English and abstract in Arabic & English.