The invention discloses a traffic image splicing method for unmanned driving. The method comprises the steps of obtaining traffic images to be spliced; Constructing a Hessian matrix, and generating all interest points for feature extraction; Constructing a scale space; Positioning feature points; performing Feature point main direction distribution; Generating a feature point descriptor; Realizingfeature point matching through bidirectional matching and similar wavelet feature matching; Randomly extracting four pieces of sample data from the data set to calculate a transformation matrix H; Calculating a projection error, judging the projection error, outputting a matching point if the projection error meets the requirement, otherwise, returning, and splicing the images according to the feature points. Compared with a traditional Surf algorithm, the bidirectional similar wavelet feature matching surf algorithm has the advantages that the coarse matching precision is improved; meanwhile, the workload of next Ransac mismatching removal is also reduced, and compared with a traditional Ransac algorithm, the self-adaptive low-iteration Ransac algorithm achieves self-adaptation of the minimum matching point threshold value, reduces the number of times of iteration and improves the efficiency of the algorithm.