YOLOV8 Object Detection - River/Waterwaste Detection
Hey everyone, we’re developing a floating waste detection project using YOLOv8 trained on Roboflow with a Raspberry Pi 5 and Raspberry Pi Camera.
Right now, our model can detect trash objects, but it doesn’t properly track them when they move or re-enter the frame.
We wanted to ask for advice on:
Best way to add real-time object tracking to YOLOv8 on Raspberry Pi 5.
Whether ByteTrack, DeepSORT, or another tracker is better for lightweight embedded systems.
Tips to improve FPS and tracking stability on Raspberry Pi 5.
Whether segmentation is better than normal object detection for floating river waste.
Best practices for creating a high-quality dataset
Tips to improve mAP50-95 to around 95% or higher.
Whether recording videos of floating trash in a pool and extracting frames/images for training is a good approach.
How to avoid getting too many similar images from video frames
Recommended augmentations or preprocessing techniques for water environments.