Loading...

Crowd Counter

Deep Learning based image processing model to estimate crowd strength

Problem Statement

Vadodara Marathon, takes place every year in Vadodara where around 100,000 participants run across city. The even is hosted by Jyoti Ltd. We were asked to develop a software to help calculate the corwd strength at specific locations while the marathon is taking place.

Client Jyoti Ltd.
Role Complete deployment
Year Started 2019
Github URL Visit Website
Image
Image
Image

Software

Pytorch was used on CUDA, we trained our CNN model on Shanhai Dataset which had numerous images of crowds ranging from street to stadium full of people. The images were annotated in Matlab files consisting of the 'head' counts of every image.

Hardware

Using iot-based wifi cameras, we connected it to Raspberri Pi where the images were supposed to be stored. Applying the image processing post event, statically.

Image

Technologies used and skills learned were:

  • Python3
  • PyTorch
  • Raspberri Pi
  • Image Processing
  • IOT (connecting cameras over local servers)
  • CUDA