README.md

Visual Traffic Surveillance and Analytics System

Introduction

A complete system for the analysis of Indian Traffic Patterns

Provides a novel approach to acquiring traffic counts and link speeds

Prerequisite Installation

0. The code was tested on Ubuntu 18.04
1. First it is suggested to have your nvidia-drivers to be installed properly This can be checked by:

   nvidia-smi

2. Install cuda-10.0 and cudnn (latest for cuda-10.0)

   nvcc -V

3. Create a virtualenvironment using virtualenv and virtualenvwrapper

   pip install virtualenv virtualenvwrapper
   mkvirtualenv all
   workon all

4. Install opencv-3.4.4 (pyimagesearch website tutorial) While compiling OpenCV use this command:

   cmake -D CMAKE_BUILD_TYPE=RELEASE \
   -D CMAKE_INSTALL_PREFIX=/usr/local \
   -D WITH_LIBV4L=OFF \
   -D WITH_FFMPEG=ON \
   -D WITH_CUDA=OFF \
   -D INSTALL_PYTHON_EXAMPLES=ON \
   -D OPENCV_EXTRA_MODULES_PATH=~/opencv_contrib/modules \
   -D OPENCV_ENABLE_NONFREE=ON \
   -D BUILD_EXAMPLES=ON ..

   To test opencv for python :-

   # In python3
   import cv2
   cv2.__version__

5. Install tensorflow-gpu==1.14.0 Does not work with tensorflow-gpu==1.15.0 (You might Engine load failure even though engine is generated correctly)

   import tensorflow as tf
   tf.__version__

6. Install pytorch (pip installation, cuda-10.0,10.1)

   import torch
   import torchvision

7. Install nvidia tensorrt 5.1.5. To check tensorrt installation

   dpkg -l | grep nvinfer

8. Install boost and all its dependencies

   sudo apt-get install libboost-all-dev

Installation

1. Create a build folder

   mkdir build && cd build

2. Compile and install

   cmake ..
   make -j8
   make install
   cd ..

Creating model

Run the following command after setting up: shell ./install/createEngine --caffemodel=./models/yolov3_traffic_final.caffemodel --prototxt=./models/yolov3_traffic_final.prototxt --input=./test/test.jpg --W=416 --H=416 --class=9

if you want batching, then run shell ./install/createEngine --caffemodel=./models/yolov3_traffic_final.caffemodel --prototxt=./models/yolov3_traffic_final.prototxt --input=./test/test.jpg --W=416 --H=416 --class=9 --batchsize=2

This will create an engine file namely yolov3_fp32.engine.

Move this engine file into the models folder

Adjust Config

NOT IMPLEMENTED YET

Run

• For RTMP streams run main.sh

   ./main.sh --engine=./models/yolov3_fp32.engine --pipe-path=/tmp/fifopipe1
   --segment-path=./segments/test_cam/
   --line-coord=./configs/line_coord/test_cam.json
   --intrinsics=./configs/cam_intrinsic/test_cam.json
   --qlength_conf=./configs/qlen_conf/coord.json
   --rtsp_strm=rtmp://aaa.bbb.cc.ddd/0

• For RTSP streams run main.sh

   ./main.sh --engine=./models/yolov3_fp32.engine --pipe-path=/tmp/fifopipe1
   --segment-path=./segments/test_cam/
   --line-coord=./configs/line_coord/test_cam.json
   --intrinsics=./configs/cam_intrinsic/test_cam.json
   --qlength_conf=./configs/qlen_conf/coord.json
   --rtsp_strm=rtsp://aaa.bbb.cc.ddd/0

Results

Soon to be out

Authors/Maintainers

• Prajwal Rao - [email protected], [email protected]
• Sadgun Srinivas - [email protected], [email protected]
• Armaan Puri - [email protected]