To enable effective human-robot collaboration, robots are expected to observe the working environment, understand human intentions, and learn skills from human demonstrations. However, existing methods often require manual human assistance to help robots learn actions or rely on human to make judgments, and robots cannot directly learn from human demonstrations. This study proposes a framework based on human collaboration demonstration and robot learning, enabling robots to learn from human-demonstrated action data and assembly image information to assist humans in assembly. The framework consists of two main elements: (1) A motion trajectory learning method based on wearable sensors that learns the collaborative action from human arm motion data. (2) An image classification method based on deep learning that learns the task context from images of the assembly and parts areas. Experiments on human-robot collaboration placement of blocks confirm the effectiveness and advantages of the method developed in this work.