Computer games are the drosophilae and universal benchmarks for artificial intelligence. In recent years, sequential imperfect information game solving has always been a frontier topic in the field of computer games research. Therefore, a comprehensive analysis of the imperfect information games solving problem in computer games is carried out. Firstly, it sorts out the milestones of the landmark breakthroughs in the field of computer games, briefly introduces four new evaluation benchmarks, summarizes three research paradigms, and deeply analyzes the challenges faced by games with imperfect information. Secondly, it focuses on investigating the game model and solution concept of a sequential imperfect information games, and briefly introduces it from three aspects: game formulation, sub-game and meta game, solution concept and evaluation. The offline strategy solving methods are systematically sorted from three perspectives of algorithmic game theory, optimization theory, and game theoretic learning. The online strategy solving methods are systematically sorted from three perspectives of opponent approximate learning, opponent discriminant adaptation, and opponent generative search. Finally, the challenges faced are analyzed from three perspectives: environment, agent (opponent) and strategy solving. The future research frontiers and prospects are given from five aspects: game dynamics and strategy space theory, multi-modal adversarial game and sequential modeling, general strategy learning and offline pretrain, opponent modeling (exploitation) and anti-exploitation, ad-hoc teamwork and zero-shot coordination. This paper provides a comprehensive overview of current imperfect information game solving, hoping to inspire related research in the field of aritificial intelligence and game theory.