TomatoFT / Vehicle-Trajectory-Prediction-in-Ho-Chi-Minh-city-streets

Predict the trajectory of the vehicles in HCM city streets with YOLOv7 + DeepSORT + CNN-LSTM/CNN-GRU.

Vehicle Trajectory Prediction in Ho Chi Minh City streets.

This is the final project for DS201 - Deep Learning for Data Science of the University of Information Technology course.

The purpose of this project is to predict the trajectory of the vehicles in Thu Duc District, Ho Chi Minh city, Vietnam. There are 3 main sub-tasks for this project. There are Vehicle Detection, Vehicle Tracking and Vehicle Trajectory Prediction.

I choose YOLOv7 for Vehicle Detection Task, DeepSORT for Vehicle Tracking Task and CNN-LSTM/CNN-GRU for Vehicle Trajectory Task.

📁 Vehicle Trajectory Training Dataset

You can access the dataset used for training the Vehicle Trajectory model by following this link: Vehicle Trajectory Dataset on Kaggle.

📓 Training Model Notebook

To explore the training model and run experiments, you can use the following Google Colab notebook: Vehicle Trajectory Training Model Notebook.

Feel free to use these resources to enhance your understanding of vehicle trajectory modeling and analysis.

Tech Stack:

Demo

city_new_4.mp4

You can see in the small demo. The center of every bounding boxes is the current state, the model is trying to predict the center point in the next state.

In this project, we use Time Series Approach to predict the trajectory of the vehicle.

Set up

Download the pretrained weight of YoloV7 in this link: https://github.com/WongKinYiu/yolov7

Go to IO_data/input/video folder to store the download or recorded of the vehicle on the street.

Go to run.py to change some hyperparamater of function tracker.track_video

Create the virtual environment

conda env create -f environment.yml

Activate the virtual environment

conda activate vehtrajpred

Run the inference file

python3 run.py