AI & ML Projects

This project is a preliminary part of an ongoing Industrial Employee Attendance Project at Fab Lab, Independent University Bangladesh. I used the simplest "face_recognition" API for python in order to recognize faces in front of the camera. Along with it, I created an Attendance project that will use the webcam to detect faces and record the attendance live in an excel sheet. I also incorporated "datetime" library for recording the real-time when attendances are listed. The project also ensures that there is no overlap of attendance because it keeps track of the existing attendance list whenever a face is recognized.

GitHub: https://github.com/MZayed47/Smart_Attendance_System

This project is a part of my ongoing research work on “Real-Time On-Road Vehicle Detection and Distance Estimation”. For this project, I have used the YOLOv4 algorithm (published in April 2020) which is considered the fastest and most accurate version of YOLO. But unlike the previous project, where I created a YOLOv3 model from scratch, in this project, I loaded the actual YOLOv4 model from its core repository and then converted the darknet model into the TensorFlow model (tf.pb). I also explored tflite.pb which is very lightweight and compatible with android phones and other edge devices. This modified TensorFlow model can detect objects from several angles: car dash camera, human-eye height and live feed from drones or helicopters.

The major improvement in YOLO v4 is, it takes the influence of state of art BoF (bag of freebies) and several BoS (bag of specials). The BoF improves the accuracy of the detector, without increasing the inference time. They only increase the training cost. On the other hand, the BoS increase the inference cost by a small amount however they significantly improve the accuracy of object detection. The further improvement of this project for my research work is to estimate the distance of the on-road vehicles

GitHub: https://github.com/MZayed47/Real_time_OD_webcam_dashcam

This project is a part of my second research work on “Real-Time Detection and Recognition of traffic signs in Bangladesh using YOLOv3 Detector”. Before getting started on that research, I had to learn the basics of computer vision algorithms and how they work. I studied several algorithms and found that YOLOv3 is the latest and faster than any other algorithm till now. So, the primary step I took before conducting the research is building this project to detect objects in images using the YOLOv3 algorithm. I used Keras library to build the YOLOv3 model from scratch and loaded it with a pre-trained weight file. Then I used both Keras and TensorFlow to load the model and detect several classified objects in images.

YOLOv3 is now highly used widely for computer vision projects because of its speed and lightweight model. I have continued my exploration in the field of Computer Vision using YOLOv3. The further improvement of this for my research work is to train the model with my team’s custom-made dataset and then run a test accordingly.

GitHub: https://github.com/MZayed47/Object_Detection_using_YOLOv3