Mina Abd El-Massih

I am a Computer Vision Engineer, an online Data science MSc student at University of East London. My work includes dealing with image/video input of CCTV camera data, creating/cleaning image and video datasets, training and fine-tuning a variety of deep learning models (YoloV4, Faster RCNN, etc...) and using trackers to solve different challenging real life projects (Accident Detection, Violence Recognition, Vehicles/People entering restricted area).

  Cairo, Egypt         Gmail: [email protected] Phone: +201284400156 

Education

2015 - 2019

Ain Shams University

Scientific Computing Department

Ranked 2nd out of 135, GPA: Very Good.
Graduation Project Grade: Excellent.


Work Experience

January 2020 - August 2022

Machine Learning | Computer Vision Engineer

Advanced Programs Trd Co.

- Working on Video Analytics projects to turn Riyadh city into a smart city.
- Training and finetuning SOTA models like YoloV4, Faster RCNN, etc..
- Working on challenging real-life projects using state-of-the art trackers.
- Dataset creating, cleaning, ..etc
- Hands-on experience with Deepstream | Nvidia TLT.
- Hands on experience with using, modifying Gstreamer pipelines in Python.

March 2019 - May 2019

Research & Development Internship

Valeo

-I was a research and development intern at Valeo, I worked on autonomous driving and participated in CARLA challenge representing Valeo in the challenge.
-The challenge was mainly about using deep learning to allow a simulated car to pass a couple of scenarios which show how good the car is at autonomous driving and how good the car passes the scenarios.

Projects

Accident Detection

Using information obtained from tracker, I was able to detect anomalies by calculating pixel velocities and direction of vehicles moving and from that information a normal range of velocities and magnitudes was generated and any outliers were considered anomalies which cause or had accidents.

Vehicle Entering Restricted Area

Implemented with Python and OpenCV, using YoloV4 and Nvidia's Deepstream 5 NvDCF tracker.

Object Dropped/Removed detection

Implemented with Python, using background subtraction SubSENSE algorithm.

Vehicle Color/Type Classification Analytics

Creating analytics based on vehicle colors and/or type, for example, a video would be created of all the red cars that pass along a certain street. The analytics are extracted from a given long video of the camera's feed or a live feed from cameras, implemented using Yolo with a tracker on Deep Stream.

Violence Recognition

The graduation Project, implemented with Python, Keras and OpenCV, a deep learning model was created in which a pretrained VGG-16 model is followed by an LSTM to extract temporal features and then it was trained on those extracted features to decide if violence existed or not. A dataset was created and added on Kaggle, can be found here.
I presented the publication about it in the presence of leading researchers and scientists. Publication can be found here.

Publications

December 2019 

Violence Recognition from videos using Deep Learning Techniques ICICIS 9th IEEE Conference

See publication.

Courses

1. Andrew NG's Deep Learning Specialization

2. Udacity's Computer Vision Nanodegree 

AI Technical Skills

Soft Skills

Languages


  • Python 
  • Anaconda | Pip | Docker 
  • ML | DL | CNN
  • Keras | Onnx | MXNet | Gluon
  • Computer Vision 
  • Github | git 
  • Linux 
  • Deepstream5 
  • Gstreamer (Python)
  • Google Cloud Platform

  • Teamwork.
  • Public speaking.
  • Problem Solver.
  • Self Learner.
  • Fast Learner.
  • Ability to work alone.
  • Self-motivated.

  • Arabic - Native
  • English - Professional
Powered By CakeResume