Browsing by Author "Dr. Rajesh Palit"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Open Access
    City Traffic Object Detection from Aerial View by Drone Using Computer Vision
    (North South University, 2022) Kazi Shehjad Islam; Md. Rafsan Khan; Md. Akib Dewan; Abdullah Al Mamoon; Dr. Rajesh Palit; 1722198042; 1812403042; 1811100042; 1621845642
    Deep learning algorithms have created a new wave of technologies that are affecting the world as we know it .From movie sites, food sites to international security, deep learning has been used in every case. As we said before, its use is very versatile. We can also use it in traffic object detection which we have tried to do in our project. Using this detection method we can count the number of vehicles on any road using its images or video footage and find the density of it. We used YOLOv5 and Faster R-CNN algorithms for doing this and we have successfully identified the vehicles on the roads. Object detection on drone podium abide a challenging assignment due to various factors such as plot of lookout, alter, barrier, balance. In dainty, large-scale drone-based A dataset boast 8,599 images (6,471 for training, 548 for validation, and 1,580) extravagantly glossed including object bounds for arduous Boxes, object categories, occlusion, truncation caliber, etc. One such algorithm is YOLOv5, developed in 2020. In this delving, every existing YOLOv5 architecture for small target detection is improved by modifying the YOLOv5 configuration. Better performance for small objects by adding a new feature fusion layer in the feature pyramid part of YOLOv5.
  • Thumbnail Image
    Item
    Open Access
    Design of a portable noninvasive Glucometer with clinical accuracy
    (North South University, 2022-06) Kazi Mosaddequr; Dr. Rajesh Palit; 1831543042
    We propose a painless Blood-Sugar level detector which is minimalistic hardware that individuals can operate regardless of their skill and knowledge. This device uses a dedicated sensor (OPT101) and light source (infra-red LED light) to collect Photoplethysmography (PPG) data from the fingertip of the subject. This data is cleaned before various features are extracted from it. Principal Component Regression and Partial Least Square Regression model are then applied to this data to find a correlation between PPG and actual Glucose level which is simultaneously collected from the subject through a commercial glucometer. The trained model is then used to test on subjects for predicting actual glucose levels using only PPG collected through the device. The results show the accuracy is on the clinical level. This device will eliminate the expensive cost of strips and needles that need to be purchased with the hardware in the traditional method. The user using this device only needs to buy the hardware, and there is no additional cost after that. Users can get an instant reading from the hardware.