In an intelligent transportation system, the vehicle detection is an important prerequisite for vehicle tracking and identification. However, the traditional vehicle detection algorithms cannot effectively maintain a balance between accuracy and real-time performance. In this paper, a new moving vehicle detection algorithm based on improved adaptive pixel segmentation is proposed. The initial background model is established based on multi frame interval image, and an evaluation method of background region change is proposed based on the spatio-temporal variation degree. Based on this, an adaptive updating strategy of learning rate is formulated. By setting a trust interval, whether the current background model needs to be updated can be adaptively determined according to the current traffic conditions and pixels whether are in the trust interval, thereby the accurate and fast detection of moving vehicles are realized. The performance indicators of Recall, Precision and F-measure of the improved adaptive pixel segmentation algorithm in different scenarios are 0.929, 0.864 and 0.888, respectively, which are higher than the traditional adaptive pixel segmentation algorithm, and the processing time of the algorithm is 88.37 ms, nearly 10 ms faster than the traditional adaptive pixel segmentation algorithm. It basically meets the requirements of high speed, high precision and high robustness of vehicle detection.