-
In this project the model with the input either as a live feed from any camera or a video file ,can detect the number of people at that instant as shown in the below video
-
This is an end to end project , python's Streamlit framework is used.
-
The works with YOLOV4 Model.
-
Down below , the sample output of a video file.
Sample.mp4
-
As the name yolo(You only look once) tells, The total prediction of image is completed in one forward propogation(one run).
-
The yolov is built with convolutional neural networks.The yolov algoritham divides the image in to specific grids of equal areas and these grids are used for detection.
-
These grids predict the co-ordinates of the grid, object label and probability.But as there are many cells predicting the main object, there will be multiple bounding boxes for the same object , to solve this problem the algoritham uses the simple non maximal suppression technique.
-
Non maximal suppression technique bascically means , it takes a look at all the bounding boxes of an object in the image and selects the the box which has the highest probability , then it removes the all the boxes which overlap with the selected box thus getting the best bounding box.
-
Yolo architecture contains 24 convolutional neural networks and to fully connected neural networks.
