The invention discloses a back sight image cognition based on-vehicle blind area early warning system and a method, wherein the system is composed of cameras (1 and 2) mounted on a rearview mirror to monitor the blind area, a digital signal processor (DSP), a sound prompting device, a power source and a plurality of data cables. The system reads the automobile travelling signal through the data cables, acquires the road surface image of the blind area (4 and 5) by adopting the rearview cameras, the processor processes the image, detects the automobile (3) in the blind area, analyzes the travelling state of the automobile, and realizes the multi-stage tracking for the automobile, when the driver intends to switch lanes, if another automobile in the blind area is not far from the automobile, then the driver is warned by the sound prompting device, so that the danger of collision during switching lanes is avoided. The back sight image cognition based on-vehicle blind area early warning system and the method are applied to the active safety driving, and has the advantages of simple system configuration, low cost, high safety performance, good real-time performance and complex environment adaptability.