With the aim to estimate depth from a pair of rectified stereo images, stereo matching plays a fundamental role in many applications including robot navigation, autonomous driving and augmented reality etc. Deep learning based stereo matching methods have achieved significant progress, however, state-of-the-art methods still have challenges in handling occlusions, textureless and repetitive regions. In addition, achieving real-time performance is critical for practical deployment yet remains another big challenge. In this project, the student will propose and implement novel methods to obtain a lightweight stereo matching network with good accuracy, which can lend well for implementation on embedded platforms with limited resources.