ARGO_Labs

This is a repository for all my work at ARGO Labs (in construction)

The goal of this project is to implement a traffic flow counter/count distinct objects for the streets of New York City (2nd Avenue Street of NYC) using ENet Neural Network (pixel wise semantic segmentation) and OpenCV.

1. Package Requisities:

   a. OpenCV 3.4.1

   b. numpy

   c. argparse

   d. imutils

   e. time

   f. cv2

2. get_street_images.ipynb - jupyter notebook to download the Google Street View (GSV) images in the images folder and then turn those images to a video (MP4 file), which is saved in the videos folder(train.mp4). Get API Key for GSV images from here

3. segment.py - python file to run the segmentation algorithm on image files

4. segment_video.py - python file to run the segmentation algorithm on the train.mp4 file and count the number of distinct objects in each frame (here each video frame is a single image as the video itself is made of several images put together) of the video. The output of the segmentation is saved as "second_avenue_output_count.avi" in the output folder.

A small clip of the output video

5. To run the segmentation on an image file:

    python segment.py --model enet-cityscapes/enet-model.net \
   
         --classes enet-cityscapes/enet-classes.txt \
   
         --colors enet-cityscapes/enet-colors.txt \
   
         --image images/image_name
   

6. To run the segmentation on the video file (mp4 in the videos folder):

    python segment_video.py --model enet-cityscapes/enet-model.net \
   
           --classes enet-cityscapes/enet-classes.txt \
   
           --colors enet-cityscapes/enet-colors.txt \
   
           --video videos/train.mp4 \
   
           --output output/second_avenue_output.avi
   

Reference: https://www.pyimagesearch.com/2018/09/03/semantic-segmentation-with-opencv-and-deep-learning