Image Identification of a Moving Object Based on an Improved Canny Edge Detection Algorithm

Edge detection plays an increasingly critical role in image process community, especially for moving object identification problems. For this case, the target object can be captured straightly via the edges beside which there is an obvious jump of grey value or texture. Nowadays, Canny operator has gained great popularity as it shows higher anti-noise performance and presents better detection accuracy in comparison with other edge detection operators like Robert’s, Sobel’s, Prewitt’s etc. However, the Gaussian filter associated with the classic Canny operator is sometimes too simple to decrease the all-type-noise. Additionally, in order to enhance the detection accuracy and lower the pseudo-edges detection ratio, two thresholds, high and low, are chosen artificially which have actually limited the adaptability of the algorithm. In this work, a compound filter, Gaussian-Median filter, is proposed to improve the smoothing effect. The self-adaptive multi-threshold Otsu algorithm is realized to determine the high/low threshold automatically according to the grey value statistic. Image moment method is conducted on basis of the detected moving object edges to locate the centroid and to compute the principal orientation. The experimental results based upon locating the edges of both static and moving objects proved the good robustness and the excellent accuracy of the proposed method.