The behavior of thunderstorms can be obtained through the monitoring of lightning flashes recorded by video camera. The flashes events shows the temporal and spatial evolution of convective cells, identifying the behavior of thunderstorms. This paper propose the use of computer vision technique to automatize the analysis of lightning flashes, currently limited by the high volume of video information generated. This system is consisted of image acquisition and processing of algorithms, including filtering, thresholding and edge detection. A two-stage process algorithm based on Sobel morphological operator is used to detect regions of interest (ROI) containing lightning flashes. Those ROI's are submitted to an edge detection stage, which returns the lightning stroke channel separated from the background image. The flash characterization is done based on the horizon line, a supervised information given to the software, and consists on classifying the lightning between cloud-to-ground flashes and intracloud flashes. The duration of each independent lightning flash is also calculated by the system. Some results of the monitoring of lightning flashes using this technique are showed in this paper too. Seven video cameras were installed to record continuously lightning flashes events on summer time at Sao Paulo, Brazil. The cameras were disposed in a 360º loop, recording all events with a time resolution of 33ms. In this period, several convective storms were recorded and analyzed.