Chicken Segmentation
This project is focused on developing a low-cost computer vision system for estimating the weight of poultry animals without relying on 3D depth sensors. Instead of using depth data, the approach is based on image segmentation to isolate the chickens from the background and extract useful visual features for weight estimation. The system is designed to capture images from three different camera angles, which are then combined and used as input to a lightweight CNN model. To make this effective, the current phase of the project is centered on building a reliable segmentation pipeline. I have been involved in preparing and organizing the dataset for training the segmentation model, which will be used to accurately separate the chickens from irrelevant background elements such as the platform, waste, and surrounding environment. The goal of this segmentation step is to generate clean masks from each camera view so that only the relevant regions (the chickens) are fed into the model, improving accuracy while keeping the system computationally efficient. This is especially important given the low-cost constraint and the need for a lightweight model that can run in practical settings. At the moment, the project is still in progress, with initial model training underway to validate the segmentation approach before moving on to full weight estimation.