Sports analytics is rapidly evolving today through the use of computer vision systems that automatically extract huge amount of information inherently present in multimedia data without much human assistance. This information can facilitate a better understanding of patterns and strategies in various sports. However, for non-professional teams, due to expense and large variations in the videos, there are no reliable systems for automatic extraction of game statistics and information. In this thesis, we consider two problems in basketball sports analytics with the goal of being robust to wide differences in video footage and being completely automatic. First, we consider the problem of parsing a game into possessions and inferring which team has possession at any time. This information provides basic statistics and the ability to easily navigate a game in terms of possessions. Second, we consider detecting shots, which allows for shot count statistics to be automatically generated. Our experiments across a wide variety of basketball video show that the approaches are accurate and robust to large differences in video type.