In order to enhance the corona detection efficiency of commercial SBUV/Visible double spectral corona detect systems and avoid the noise interference, an automatic real-time detection method detecting weak solar blind ultraviolet corona is proposed. Based on the analysis of statistical characteristics of the corona target and noise time-domain, utilizing corona characteristics in continuous time domain, the pretreatments like gray image binarization and morphological dilation are completed firstly, followed by accumulation of the N successive frames, then thresholding. At last the features of corona like location, size and so on are extracted. Upon completion of the calibration of radiation equipment, the corona's photon counting reference value can date back immediately. The mathematical model of detecting probability and false arm rate(FAR) is established. We implement the method and test on a high-speed digital processing platform TMS320DM642 using the video of corona, and the results show that its FAR is 2.85×10-5, and the processing time is less than 120 ms, indicating that the weak corona can be detected in real-time.