This project is to detect cars that move in the wrong way. Tensorflow implementation based on YOLOv3 detector and Kalman Filtering for the paper: Real-Time, Deep Learning Based Wrong Direction Detection
In this paper, we developed a real-time intelligent transportation system (ITS) to detect vehicles traveling the wrong way on the road. The concept of this wrong-way system is to detect such vehicles as soon as they enter an area covered by a single closed-circuit television (CCTV) camera. After detection, the program alerts the monitoring center and triggers a warning signal to the drivers. The developed system is based on video imaging and covers three aspects: detection, tracking, and validation. To locate a car in a video frame, we use a deep learning method known as you only look once version 3 (YOLOv3). Therefore, we used a custom dataset for training to create a deep-learning model. After estimating a car’s position, we implement linear quadratic estimation (also known as Kalman filtering) to track the detected vehicle during a certain period. Lastly, we apply an “entry-exit” algorithm to identify the car’s trajectory, achieving 91.98% accuracy in wrong-way driver detection.
- Real-time detection (up to 35 FPS on Nvidia GTX1080Ti)
- Support parallel training
- Video and data are included with annotated files.
This project is done on a Windows 10 machine with an already installed Anaconda platform. If you don't have one, you can install it from here Creating an environment can be done by running.
conda env create -f requirenments.ymlTo run this program you need to download some files from here
- Download YOLOv3 model data and put it in the model_data folder
- Download pre-trained data and put it in the model folder
- Download the test video to run the program
conda activate carThis will activate the environment created from requirenment.yml file
python object_tracking.py --input <PATH to your video file you downloaded>First, create a YOLO format list of annotations. The example of the annotation file can be seen below
Generate your annotation file and class names file.
One row for one image;
Row format: image_file_path box1 box2 ... boxN;
Box format: x_min,y_min,x_max,y_max,class_id (no space).
Here is an example:
path/to/img1.jpg 50,100,150,200,0 30,50,200,120,3 path/to/img2.jpg 120,300,250,600,2 ...
Modify the train.py file to adjust to your preference for training
python train.py
will start training
This model is reasonably fast in training the model since it uses YOLOv3 416x416 image size.
Object tracking is based on the Kalman filter algorithm and compared to the other ML algorithms it's fast and reasonably accurate, you can modify it to your own tracker.
The algorithm of estimation is simple
for the wrong direction estimation, you can modify detector_car_person.py starting from line 279 as your task and wrong direction may be different from ours
Here are the results of our work, this system is in current use, and it's practically effective.
If you think this work is useful, please give me a star!
If you find any errors or have any suggestions, please contact me (Email: u.s.saidrasul@inha.edu).
@article{Real-Time,
author = {Saidrasul Usmankhujaev, Shokhrukh Baydadaev, Kwon Jang Woo},
title = {Real-Time, Deep Learning Based Wrong Direction Detection},
year = {2020},
journal = {DOI: 10.1109/ICSET51301.2020.9265355},
}