Human detector module is an open-source perception module for the autonomous vehicle software stack. It is developed by Sandeep Kota Sai Pavan and Satyarth Praveen, the software developers at TerpBotics.
- Sandeep Kota Sai Pavan - Github Link
- Satyarth Praveen - Github Link
-
Sandeep Kota Sai Pavan (Email: skotasai@umd.edu, LinkedIn: Link): A graduate student of Masters in Engineering - Robotics at University of Maryland, College Park. I'm a robotics enthusiast interested in autonomous vehicles and deep learning.
-
Satyarth Praveen (Email: satyarth@umd.edu, LinkedIn: Link): I am currently a Robotics graduate student and a Graduate Teaching Assistant with the Department of Mathematics at the University of Maryland College Park. I was previously employed at Hi-Tech Robotic Systemz where I was involved with the perception group of the Unmanned Ground Vehicles (autonomous vehicles). I find my interest in Artificial Intelligence.
This project is the implementation of a human obstace detector for an autonomous vehicle software stack.
Pedestrian detection should be given the maximum priority in an autonomous vehicle software stack as it deals with the lives of the people. Hence, it is critical for the autonomus vehicle to predict the humans with maximum accuracy. Developing a dedicated human detector module helps in achieving improved prediction accuracies for this specific tasks without effect on the other perception modules.
An agile iterative process (AIP) is followed for the development of this project where the product backlog is developed first. Based on the priority of the tasks, the sprint cycles and tasks are decided. A daily meeting at the begging of the sprint cycle is conducted to decide on the daily tasks for each developer. The project backlog consists of the estimated time of completion which is alloted for each task. The actual time of completion is altered based on the progress of the project and the remaining tasks.
After planning the product backlog, UML flow diagram and UML class diagram for the software are created. A set of unit tests are used to verify the performance on a wide range of exple scenarios. Stub implementations are written with the functions to ensure that the code coverage of the softwrae is maintained.
The link for the product backlog, time log, error log and release log can be found here - link
The link for the sprint review document can be found here - link
This software is released under the BSD 3-Clause license.
This work is stil under development and is expected to be completed in another week.