Research on Human Pose Estimation Model Based on Convolutional Neural Network

Human pose estimation is a key area in computer vision with applications in human-computer interaction, intelligent monitoring, and motion analysis. Pose estimation aims to automatically identify human body key points in images, which has significant practical value. Recent advances in convolutional neural networks (CNNs) have made deep learning the dominant approach for pose estimation. This paper proposes a simplified CNN-based model for human pose estimation. Unlike traditional complex architectures, our model improves computational efficiency while retaining strong learning ability. To evaluate its performance, we created a random pose dataset with 100 samples, each consisting of a zero image and 17 randomly generated key point coordinates, normalized in the 0, 1 range. The model is trained using mean square error (MSE) as the loss function, and the Adam optimizer for weight updates. Through backpropagation, the model learns to minimize the gap between predicted and real key points. Experimental results show that, despite the dataset's simplicity, the model effectively locates key points and demonstrates strong learning capabilities. In the future, we plan to use more complex real-world datasets and optimize the network structure to improve performance in more challenging scenarios. In conclusion, the proposed CNN-based model demonstrates its potential for posture estimation tasks.