The invention relates to an automatic detection method for traffic accidents based on collision sound. The collected sound signals are divided into frames, and the total energy distribution in the time domain is detected first. Processing of one frame; next, judge the energy distribution in the frequency domain, use wavelet to extract the characteristic components of the energy distribution and judge, if the collision condition is met, further recognition will be carried out, otherwise, it will directly jump out and wait for the processing of the next frame; finally, perform ACF detection , if the condition of collision is met, it is judged as a collision sound, otherwise, it waits for the processing of the next frame. The invention can suppress the calculation amount to the minimum while improving the detection accuracy rate and reducing the false alarm rate.