The traffic data is collected using some sensors on the road but often some data is missing. This missing data can be completed using some data imputation algorithms. The traditional data imputation algorithms, which are generally based on probabilistic and regression methods, often fail to predict data in case of higher missing ratio. This paper introduces a novel method that transforms traffic data into spatio-temporal images. The capsule neural network is used on these images to reconstruct missing data accurately. By treating the data as an image, the capsule network identifies patterns that traditional methods miss. The experiment conducted on the traffic data provided by the Caltrans PeMS showed that the capsule network improved the imputation accuracy and offered stable error distribution. It lowers the RMSE rate to approximately 5.00, MAE to approximately 4.00, and MRE to approximately 0.1 in the case of a 50% missing data ratio.