To achieve precise recognition of sports fouls, research is being conducted on image feature mining based methods for identifying sports fouls. Firstly, use binocular vision technology to obtain sports action images, and use block matching algorithm to complete stereo matching processing of the images. Then, the visual background extraction (ViBe) algorithm is used to perform foreground detection on sports depth images. Finally, after mining image features, the mining results are input into a dual stream convolutional neural network model to establish an intelligent recognition model for sports foul actions, and use this model to achieve the recognition of foul actions. Through testing, it is known that the new method can achieve stereo correction of sports action images and accurately identify sports foul actions in the presence of background disturbance.