Abstract: In this article,an algorithm for vehicle detection is proposed using the improved histogramof oriented gradients(HOG)features,in which a fusion of weighted multi-channel HOG features isemployed.First,the color image is enhanced by the method based on the mean of brightness.Afterthat the image is remapped into HSV color space.In consequence,the multi-channel HOG feature isobtained by combining the adaptive-weight HOG feature for each channel.Finally,the support vectormachine(SVM)is trained by such feature to detect the vehicles.Experiments prove the robustnessand precision of this algorithm and it reaches the requirement for all-weather vehicles detecting interms of the one using the traditional HOG features.