According to the requirements of elevator-group-control systems, a new method based on computer vision detection technology was proposed to obtain the number of passengers waiting outside of an elevator. As the detecting target of the system was the passengers waiting outside of the elevator, the feelings of passengers, building styles, the installation angle of a camera and the complex background would effect the detection pattern. Therefore, this paper took the human head as the model to implement the pattern recognition and to detect the number of passengers waiting outside of the elevator. The proposed method based on computer vision detection algorithm combined Mean Shift image segmentation and Support Vector Machine (SVM) classification and recognized the human head features according to the angles of image acquisition and special lenses of cameras. It can obtain accurate recognition results. Experimental results show that the method has an image processing speed by 2 s/image in real time and the accuracy above 80%, which meets the needs of elevator-group-control systems. As a results, the transport efficiency has been improved greatly due to the stable input parameters for the elevator-group-control systems.