The invention discloses a video processing-based intelligent signal lamp state detection method and a detection system. For the video of a signal lamp, the regional position of the signal lamp is determined based on the positioning algorithm. At the same time, the imaging quality of the detection region of the signal lamp is analyzed. Therefore, whether the video quality of the signal lamp is abnormal or not is judged, and then the abnormal level of the video quality is given. If the abnormal level of the video quality is within the range of the abnormal level, the region of the signal lamp is segmented out in combination with a Gaussian model and an HSI color space. Meanwhile, the connected domains of the detection region of the signal lamp, that meet the chromacity requirement, the brightness requirement and the saturability requirement, are extracted based on the clustering analysis. After that, an optimal connected domain is extracted based on the geometric feature analysis on shape, color, position and the like. In this way, the state of the signal lamp can be determined. At the same time, the shape of the signal lamp is recognized by an SVM classifier. According to the invention, no hardware accessory and the installation thereof are required. The method and the system are stable, effective, strong in self-adaptability and small in complexity, which are capable of monitoring the state of the signal lamp in real time. Meanwhile, in harsh environments and poorly imaged conditions, the method and the system provide more accurate and more effective law enforcement evidences for police traffic management.