In order to solve the problem of automobile fuel tank cover positioning in natural scene, a visual detection method of automobile fuel tank cover based on HoG and SVM in unstructured environment was proposed.The car image was preprocessed and the dark detail features were extracted by using multi-scale bottom-hat transform; the improved maximum entropy threshold segmentation method was used for image segmentation; the connected region labeling method was used to make statistics of the binary image, and the target candidate regions were determined in the original image; the HoG feature and support vector machine were used to classify and decide the candidate area to locate the fuel tank cover.The results show that by using the method, the fuel tank cover position can be accurately detected, even if the image has uneven illumination, dust on the surface of the car coverings and fuzzy details, it also has a good positioning effect.