Progress Log of experiments and findings
Researching methods to detect the sharpness of a road bend on approach, and determining the risk relative to vehicle speed.
Initial approach was to train an RNN to handle the entire problem; however, this would require a massive dataset and hours of compute, and presented too large of a challenge to achieve in the given time frame.
Therefore, I am using traditional/mathematical computer vision approaches to clean, analyse, and extract relevant features, which can then be processed either in a neural network or algorithmically to detect and label the sharpness of a bend.
https://github.com/AAP9002/Third-Year-Project
Tickets are planned on JIRA and assigned to an epic, which relates to a stage of the projects development. I have connected this to git hub to encourage:
---
config:
theme: default
---
%%{init: { 'gantt': {'leftPadding': 250, 'bottomPadding': 100 } ,'themeCSS':
'rect[id^=vert] { height: 100%;text-anchor: middle; transform: translate(9px, 47px); y: 0; width: 1.5px; stroke: none; fill: red; } text[id^=vert] { fill: red; y: 100%; transform: translate(-0px, 0px); font-size: 15px;}'
}}%%
gantt
todayMarker off
title Project Gantt Chart: Road Bend Classification
dateFormat YYYY-MM-DD
section Deadlines
Code Submission :milestone, vert, 2025-03-21, 0d
Final Submission:milestone, vert, 2025-05-05, 0d
section Research & Planning
Research Existing Approaches, Implementations and Limitations :done, m0a, 2024-09-01, 180d
Planning Processing Pipeline :done, m0b, 2024-9-15, 10d
Stereo Camera System :done, m0a, 2024-9-25, 10d
section Milestone 1:Data Preparation
Data Exploration and Preprocessing : m1a, 2024-010-01, 30d
Bend Detection and Labelling Experiments: m1b, after m2a, 70d
section Milestone 2: R-VP Detection
Feature Extraction and Matching : m2a, after m1a, 15d
R-VP Estimation Experiments :m2b, after m2a, 55d
section Milestone 3: Dataset Generation
Dense Optical Flow Experiments:m3a,2024-12-30, 25d
Combine Bend and R-VP Data :m3b, after m3a, 10d
Dataset Generation and Cleaning :m3c, after m3b, 20d
section Milestone 4: Deep Learning Model Development
Model Architecture Design :m4a, 2025-02-20, 15d
Model Training and Evaluation :m4b, 2025-03-01, 30d
section Project Evaluation
Report Planning and Critical Reflection :crit, m5c, 2025-03-10, 7d
Final Report Writing :crit, m5b, 2025-03-12, 54d
Screencast :crit, m5c, 2025-04-28, 7d
This initial pipeline was made after the first R-VP method (R-VP through Hough lines).
I have broken the project into 3 parts that can be explored concurrently
For part 3: I plan to use the ground truth data from part 2, so I can get started on research and implementation