Advanced robot navigation and pathplanning ( Assignment 2 in Computer science and robotic engineering)
Task required previously designed Arduino robot to recognise traffic signs on the camera and perform movement either on left or right
For this purpose advanced machine learning algorithms were used to train neural network to clasify and valide traffic signs
Trained model was integrated in python script with OpenCV to process required task on live camera
Script was integrated in Robotic Operating System (ROS) as mainframe development and communication platform of the Arduino designed robot
Video presentation of performance with description can be reviewed with link below:
https://drive.google.com/drive/folders/1VLR-XyFgT5LPZoJVJFnIOW50bDpxL9nQ?usp=sharing
More detailed description can be reviewed in Assignment2...docx file
Part 1 - Neural network, python and OpenCV
Initially core component of neural network was to provide relevant dataset for high accuracy results
For this purpose German Traffic Sign Recognition Benchmark (GTSRB) were selected containing 35000 images
of road signs within 44 classes -
https://benchmark.ini.rub.de
Code was developed in python 3.6.2 version with essential libraries as OpenCV,Sklearn,Tensorflow and others
Full requirements and in depth explanation can be reviewed on
https://www.computervision.zone/courses/traffic-sign-classification/
Values provided in my script (pithontrain.py) have 98% stable performance in classification and validation, however it was 30min training process on powerfull desktop station
I was using TensorFlow "save" function in my solution as pickle library was depricated on python3 versions and higher to
generate file model.5h
This file contains attributes and values from neural network training results in binary form and used to seperate OpenCV script from training one
Since running training part each time using the camera would be ineffective
traffictest.py is the main script used with USB camera and OpenCV library. Previously generated model.5h is being loaded and used for validation and classification of the road sign image on the camera
Part 2 - Linux, ROS and Arduino
ROS requires Linux based operating system - for this purpose I was using Ubuntu 18.03 version.
https://ubuntu.com
Windows users might consider option to use VMworkstation16 as option to have Linux virtual desktop station installed on the Windows OS.
ROS melodic distribution version was installed as main frame platform of robot development
http://wiki.ros.org/melodic/Installation/Ubuntu
To perform communication between traffictest.py script in ROS and Arduino robot, script was modified to contain sending messages of road signs on which robot will respond with his programmed code to respond. In my case it were 4 classes - Turn_left, Turn_right, Forward and Stop.
Final version of modifed script located in SRC folder - loadfinal.py
Within other scripts of publisher and subscriber topics.
ROS_TopicF.ino.ino contains my configuration of Arduino designed robot. Components and pin configuration might differ, however essential part goes of integration of stg_messages in Arduino code as main bridge between ROS and Robot.
