Sequential images provided by optical cameras are ideal for realizing high-precision autonomous navigation for planetary landings, but due to limited computational resources, a large amount of image information cannot be processed on deep space probes, and the rich landmark features in the sequential images must be selected. The contribution of landmark features to navigation accuracy is usually measured in terms of observability degrees, but the traditional method only observes the locally optimal landmarks at a single moment, and the navigation accuracy will be affected by the gradual decrease of observability degrees when observing the same landmark several times in a row before landmarks are reoptimized. In this paper, we firstly establish a model of planetary landing segment sequential image autonomous navigation system, construct a sequential image observability degree for multiple observations, which guides the selection of the landmark with the highest observability degree at multiple observations. After mathematical simulation, the landmark selection based on the sequential image observability degree is verified to have higher navigation accuracy than the traditional single-moment selection method.