The invention provides a real-time traffic light recognition method based on space-time correlation and priori knowledge, and belongs to the field of traffic information detection in the intelligent transportation industry. The method includes the steps that firstly, regions of interest are positioned on an original image through the priori knowledge, and the regions unrelated to a traffic light are filtered out through empirical values; secondly, the red region and the green region of the traffic light are extracted and filtered on this basis through shape features; thirdly, sub-regions obtained through filtering are read in, the HOG features of the sub-regions are sequentially extracted, and a traffic light sample is trained through a classifier; fourthly, the current traffic light is recognized according to a discrimination function of the classifier, wherein if the front light is green, driving can be achieved, if the front light is red, a parking signal is sent out, and if both the green front light and the red front light exist, whether driving can be achieved or not is determined according to the space-time correlation information and lanes where vehicles are located. The method conforms to the detection and recognition characteristics of the traffic light, information of the traffic light can be accurately detected in real time, and the method is used in an intelligent vehicle and assists in correct and safe driving of the intelligent vehicle.