In the outdoor working environment of inspection robots, recognizing the number of meter was susceptible to illumination. An adaptive adjustment algorithm for meter images based on 2D-Gamma function was studied. Then an algorithm based on Maximally Stable Extremal Region (MSER) was proposed to extract the pointer. Firstly, the reflection component was extracted by three-scale Gaussian functions, 2D Gamma function was constructed to automatically adjust brightness of the reflected or overshadowed region of image. Secondly, the pointer region was extracted through two MSER detections. Thirdly, on the condition that the pointer passed through the axis of dial, the pointer was precisely positioned by thinning algorithm and Progressive Probabilistic Hough Transform (PPHT) to improve the accuracy of positioning lines. Finally, on basis of comparing the positions of two endpoints by PPHT with axis, the direction of pointer was directly determined, thus calculating the number was more convenient. The experimental results show that the proposed method can deal with different types of meters under different lighting conditions. Moreover, the correct rate of identification reaches over 94%.

Concerning the credibility problem of the classification results in face recognition, a face recognition method based on the theory of uncertainty was proposed. Firstly, in order to estimate 3D features, Active Appearance Model (AAM) and triangulation were used to process two 2D images of unknown object. Then, the score of each object in the database was estimated, and two images were further processed through uncertainty. Finally, the decision was made based on the estimated scores and the estimated uncertainty classification list. All identified objects and their corresponding credibilities were stored in the classification list. Stereo vision system with two cameras captures face images of various postures in the experiment. Compared with a similar probability forecasting measurement method, the correct detection rate of the proposed method was increased by 10%, and the false detection rate was reduced by at least 9%. The experimental results show that the classification accuracy is improved by constructing the uncertainty information of 3D image feature and adopting appropriate statistical method.