Abstract We propose a novel method for extracting target objects such as pedestrians from a given rough initial region of infrared video images from automobiles. The detection of pedestrians from cars is imperative for predicting their subsequent movements in safe driving assistance. However, the automatic extraction of pedestrian regions with various clothing styles and postures is difficult due to the complexities of textural/brightness patterns in clothing and against background scenes. We approach this problem by introducing an object extraction method that reveals the object boundaries while assimilating the variation of brightness distribution. The proposed method can correctly extract the object area by introducing a brightness histogram as a probability density function and an object shape distribution map as a priori probability for the preprocess of the kernel density estimator. We first confirm that the accuracy and computation speed in general cases are improved in comparison to those of the original extractor. We also extend the object distribution map to reflect the shapes of pedestrians with various postures in real situations. Preliminary experimental results are given to show the potential of our proposed method for extracting pedestrians in various clothing, postures, and backgrounds.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Extraction of Pedestrian Regions Using Histogram and Locally Estimated Feature Distribution

    Mase, K. / Imaeda, K. / Shiraki, N. et al. | British Library Conference Proceedings | 2009


    Pedestrian recognition using stereo vision and Histogram of Oriented Gradients

    Toya, Ayato / Hu, Zhencheng / Yoshida, Takehumi et al. | IEEE | 2008



    Combination of Feature Extraction Methods for SVM Pedestrian Detection

    Alonso, I.P. / Llorca, D.F. / Sotelo, M.A. et al. | IEEE | 2007


    Combination of feature extraction methods for SVM pedestrian detection

    Alonso, I.P. / Llorca, D.F. / Sotelo, M.A. et al. | Tema Archiv | 2007