Improving the Visual System of Humanoid Robots with an Enhanced YOLOv8 Algorithm

To further enhance the target recognition capabilities of humanoid robot vision systems, an improved the Visual system of humanoid robots with an enhanced YOLOv8 algorithm is proposed. Firstly, a scale attention structure is introduced to replace the C2f module in the original network, thereby expanding the receptive field and strengthening its feature extraction capabilities. Secondly, selected convolutional units in the original network are eliminated to reduce the algorithm's complexity. Lastly, the loss function of the original network is further enhanced during training to improve the algorithm's stability in detecting diverse targets. Experimental results indicate that the proposed algorithm reduces the number of parameters by 5% compared to the original YOLOv8 network, and outperforms comparative methods on both standard datasets and in real-world environments, making it more suitable for use in humanoid robot vision systems.