Through the preliminary simulation and the Matlab simulation experiment based on the color and shape of the traffic sign detection method, the traffic mark detection method based on machine learning is proposed and realized. Because the traffic signs in the natural scene often have complex background, the traffic detection method based on machine learning can more accurately identify the traffic signs under the natural scene. Through the Android NDK technology, the C ++ language image processing program can be crosscompiled into the dynamic link library, which can be called in the Android side by the Java program to solve the problem that the commonly used computer vision library on the Android platform provides incomplete interfaces and incomplete data. The road traffic sign recognition algorithm is applied to the Android system platform, and the related function test is carried out.