{"id":9429,"date":"2024-10-27T17:50:06","date_gmt":"2024-10-27T16:50:06","guid":{"rendered":"https:\/\/dronesonen.usn.no\/?p=9429"},"modified":"2024-10-28T12:34:22","modified_gmt":"2024-10-28T11:34:22","slug":"toyzrgone-week-6-7","status":"publish","type":"post","link":"https:\/\/dronesonen.usn.no\/?p=9429","title":{"rendered":"ToyzRgone \u2013 Week 6 & 7"},"content":{"rendered":"\n
\n
\n

Welcome to the combined sixth and seventh ToyzRgone project blog!<\/strong> <\/p>\n\n\n\n

Due to a large portion of the team is focusing on resit exams and balancing commitments to other courses, we decided to merge these two weekly updates into one. We\u2019ve made steady progress toward our goal of building a fully functional self-driving robot. Below, you\u2019ll find the key highlights from our recent work and the steps we\u2019ve taken to keep moving forward.<\/p>\n\n\n\n

Hamsa Hashi<\/strong> \ud83d\udcbb<\/h2>\n\n\n\n
\n
\n

What Was Done<\/strong><\/p>\n\n\n\n

The past two weeks, despite exam preparations, Sokaina and I focused on improving the dataset and fine-tuning the model. Last week\u2019s testing showed the need for more varied images to distinguish target objects from other round objects. Therefore, we expanded the dataset and set up a new testing process for more accurate evaluation.<\/p>\n\n\n\n

\n
\n
\"\"
The file structure I have used for training my custom dataset with YOLOv8. https:\/\/harshavardhanyellanki.github.io\/YOLOV8_GUI\/cfg\/<\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n

<\/p>\n\n\n\n

The previous dataset and file structure only included training(train) and validation(val) sets, which limited the model’s ability to properly test whether it had genuinely “learned” throughout the epochs. To address this, I have now added a test<\/strong> set with images, following best practices for creating a custom dataset based on the documentation. This adjustment will allow for a more accurate assessment of the model’s performance on unseen data.<\/p>\n<\/div>\n<\/div>\n\n\n\n

Sokaina and I took additional pictures of the blue, red, and green balls in realistic environments, using a Python script she created that connects her phone to Google Drive for automatic uploads. Great job from Sokaina! both efficient and a real time-saver, in my opinion. We also added images of other round objects to train the model to ignore them. A separate test set was created with images not included in the training set to ensure the model is truly learning to recognize objects.<\/p>\n\n\n\n

Fine-Tuning Training Parameters<\/strong><\/p>\n\n\n\n

\n
\n
# Tren YOLOv8-modellen med optimaliserte parametere\nresults = model.train(\n    data=os.path.join('\/content\/drive\/MyDrive\/ComputerVisionEngineer\/ObjectDetectionYolov8GoogleColab', \"google_colab_config.yaml\"),\n    epochs=50, imgsz=640, lr0=0.01, momentum=0.937, weight_decay=0.0005, augment=True, save_dir=train_results_dir\n)\n<\/code><\/pre>\n<\/div>\n<\/div>\n\n\n\n
Before the new training, We reviewed Ultralytics documentation and adjusted parameters to better balance accuracy and performance. The image size was set to imgsz=640, with a learning rate of lr0=0.01 and momentum=0.937 for smoother training. To prevent overfitting, I used weight_decay=0.0005, and enabled augment=True to make the model more robust.<\/p>\n\n\n\n
Challenges with Testing and Hardware<\/strong><\/p>\n\n\n\n
Despite improvements, we think the model still struggled with precision under poor lighting, and the mAP results were not fully satisfactory. The Raspberry Pi 4 also faced performance issues with real-time detection, and camera stream delays made testing challenging. As a result, I moved development to my Windows machine, which offers better hardware performance for real-time detection and model testing.<\/p>\n\n\n\n
\n
\n
\"\"
Part of code for Object detection using our own custom dataset <\/strong><\/figcaption><\/figure>\n<\/div>\n\n\n\n
\n
\n
\n
\"\"<\/figure>\n<\/div>\n\n\n\n
\n
\"\"<\/figure>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n\n\n\n
Moving forward, I plan to handle the more complex tasks, such as object detection, on my Windows machine to bypass the Raspberry Pi’s limitations. I will subscribe to the camera feed from the Raspberry Pi, ensuring we don’t compromise performance while still using the Pi’s camera capabilities.<\/p>\n\n\n\n
Next week<\/strong><\/p>\n\n\n\n
In the coming week, I will also begin working on the robot’s movement system<\/strong> together with Philip. We will start by researching NEMA 17 stepper motors and how they can be integrated with the camera system. To begin, I\u2019ll use a game controller to test the basic functionality and ensure everything works as expected. Once confirmed, the stepper motors will receive movement commands based on the camera\u2019s x\/y coordinates. The long-term goal is to integrate LiDAR with ROS to enable SLAM (Simultaneous Localization and Mapping), allowing us to capture x, y, and z coordinates for objects the robot needs to pick up.<\/p>\n\n\n\n
<\/p>\n\n\n\n
Sokaina Cherkane<\/strong> \ud83d\udcbb<\/h2>\n\n\n\n
NB. Exams on both week 5 and 6. <\/strong><\/p>\n\n\n\n
 To be more productive, Hamsa and I agreed to work on different datasets. I will train the one that I made from scratch on VS Code. Meanwhile, he will train an imported one from Roboflow directly to the Raspberrypie. <\/p>\n\n\n\n
Both of us met some errors and challenges, where the dataset wasn\u2019t well trained. (check the pictures below)<\/p>\n\n\n\n
   <\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
The program is not precise since it mixes different colors. 1 is <\/strong>meant to be red, 2 is blue and 4 is green. <\/strong>But the program is still inaccurate. Plus, the camera on the RP couldn\u2019t distinguish a human from a ball. We came to the conclusion that we must combine both his and my dataset, re-train it, and test it directly on the RP.<\/p>\n\n\n\n
Uke 7. <\/strong><\/p>\n\n\n\n
\n
\n