Top-View Noise-Filtered Point Cloud of 2-wheelers, 4-wheelers and Pedestrians

Citation Author(s):
IIIT Hyderabad
Submitted by:
Pranjal Mahajan
Last updated:
Tue, 04/09/2024 - 16:51
Data Format:
0 ratings - Please login to submit your rating.


Various modes of transportation traverse our roadways, highlighting the importance of object classification for improving traffic safety. Optical sensors that rely on visual data encounter challenges in adverse weather conditions, where poor visibility hinders target classification. In this project we use an off-the-shelf millimeter wave Frequency Modulated Continuous Wave (FMCW) radar -- Texas Instruments IWR1843BOOST module to classify on road objects. By combining the radar module, Robot Operating System (ROS), and Python scripts, we extracted a dataset of 3D point cloud images. The images were preprocessed to create top-view noise-filtered images, and using Machine Learning (ML) models, they were classified into 2-wheelers, 4-wheelers, and pedestrians. The ML model was trained on a dataset comprising approximately 15,000 images.


1. Download and unzip the dataset file.

2. You will find 3 folders inside it. 1) Humans, 2) 2-Wheelers, 3) 4-Wheelers

3. Inside the three folders you will find folders corresponding to different types of cars, bikes and humans.

4. You will also find .Numpy files from which all these point cloud images were generated.

5. Folder naming format: "DateMonth_Category_CategoryName_CXX_DistanceFromRadar". Example: 21May_Bike_Hero_Glamour_C15_2.8m where 21May is the DateMonth, Bike is the category, Hero_Glamour is the category name, C15 is the CFAR setting of radar, 2.8m is the distance from radar.

5. All in all there are a total of 15000 images approximately.

6. For the file naming convention: CXX means the CFAR of radar is set to XX value. Ex. C15 means the CFAR is set to 15.



Funding Agency: 
I-HUB DATA at IIIT Hyderabad