Research On Object Detection And Location Algorithm Based On Image Gradient And Curvature Information Constraints

The computer technology has been drastically developed since the20th century. Ithas been applied in a wide rage of people’s life. Then people begin to explore to make useof the computer instead of the human’s brain to distinguish objects, so the artificialintelligence technology begins to rise. As one of the most important subjects in the field ofartificial intelligence, object recognition has been got more and more attention by the mostscholars in the world, it has been deeply researched, and the scholars do their best to try tofind a more effective and simple feature description algorithm in order to improveaccuracy of object detection. But there are some drawbacks of the popular featuredescription algorithm, such as higher feature dimension, higher computational cost and soon. Therefore, the paper mainly researches on the gradient and curvature information ofthe image based on the latest research results of the word, applies them into the imagefeatures describing problems, so as to improve the performance of the feature descriptionalgorithm.Firstly, research image gradient information. The paper applies the gradientinformation and edge detection to build a new edge map as a pre-treatment process ofHough transform, and then detects straight lines. This method can quickly and accuratelydetect the straight lines in the image with low computational cost, less storage space andcleaner Hough space. Experimental results show the good performance of this method.Secondly, research image curvature information. In reality, the image is oftenconstituted by the curve. This paper focuses on how to make better use of the curvatureinformation to detect the object. The paper explores to apply the isophote curvature ofcircle to map the circle’s center and roughly estimate the radius. So the method is calledthe multi-round detection algorithm based on curvature constraints. Experimental resultsshow the feasibility and effectiveness of this method.Finally, research the feature description algorithm combined with gradientinformation and curvature information. This paper proposes new methods that make use ofgradient information and curvature information to construct the histogram combination with HOG algorithm to form improved HOG algorithm. The improved HOG algorithmcan successfully detect object more accurately from the image and increase the robustness.The experimental results confirm the feasibility and good performance of this method.