基于OTSU分割和HOG特征的行人检测与跟踪方法

传统的HOG算法针对整幅图像进行行人特征提取,大量的非人窗口计算必然降低检测的准确率和效率。为此,提出一种基于OTSU分割和HOG特征的行人检测与跟踪方法。利用OTSU算法以最佳阈值分割图像,在分割区域的基础上进行Canny边缘检测,通过边缘的对称性计算确定行人候选区,继而采用经PCA方法降维后的HOG特征和隐马尔可夫模型对行人候选区进行检测验证。最后,以确定的行人区域为跟踪窗口,利用CamShift算法跟踪行人。多组实验结果证明,本文方法的行人检测效率和精度均有所提高,跟踪性能稳定、可靠。

Pedestrian detection and tracking method based on OTSU segmentation and HOG feature

The traditional HOG algorithm extracts pedestrian features from whole image, a large number of non-human window calculation is bound to reduce the accuracy and efficiency of detection. In this connection, a pedestrian detection and tracking method based on OTSU segmentation and HOG feature was proposed. The image was segmented by the OTSU algorithm with the best threshold value, on the basis of the segmentation region, the image contour could be generated through canny edge detection, and methods of applying symmetry to calculate the image edge could determine human candidate region. Then combining HOG features after PCA dimensionality reduction with Hidden Markov Model to detect and verify pedestrian candidate region. Finally, taking determined pedestrian area as the tracking window to complete tracking pedestrian by using CamShift algorithm. Several experiments results prove that the efficiency and accuracy of pedestrian detection were improved by the method of this paper, and its tracking performance was stable and reliable.