Basketball, as an offensive and defensive game centered around high altitude, has become an international mass competitive sport. Traditional methods cannot comprehensively evaluate the future potential of players, nor can they simply add up individual competitive abilities to judge the overall competitive performance of a team. To address these issues, this article proposes a video-based RBF neural network competitive scoring method, which analyzes players' past sports behavior, captures every subtle difference in their abilities, and achieves objective evaluation of players' competitive performance. Through comparative experiments, the accuracy of the test results is improved by about 5% compared to conventional RBF methods. This indicates that the improved RBF neural network designed in this article has significantly better prediction performance than traditional convolutional neural networks. This study provides a new method for evaluating the competitive performance of basketball players and has important guiding significance for basketball training and skill enhancement.