The United Nations General Assembly resolution of September 25, 2015, defined measures aimed at sustainable, harmonious development and ensuring stability and protection of state interests. The UN Sustainable Development Goals are consistent with national and border security issues. This prominent place belongs to optoelectronic surveillance systems in state border protection. These systems are key in modern border protection strategies, providing increased efficiency, security, and reliability. One of the technologies widely used in optoelectronic surveillance systems is image contour detection. The paper proposed and researched an improved approach in gradient methods for detecting contours of images in optical-electronic surveillance systems, which consists of matching convolutional kernels for detecting contours of objects. It is shown that the dynamic adaptation of parameters of convolutional kernels, based on local orientation and normalization, makes it possible to increase the accuracy of detecting contours of objects in images in a wide range of image quality. An algorithm for adapting classical gradient methods, particularly the Prewitt and Sobel methods, is proposed for dynamic convolutional kernels. In the example of test images, it is shown that such a modification of these methods makes it possible to improve the quality of contour detection and reduces the error in the detection of weak or blurred contours with a slight increase in computational costs. The presented results of the experiments demonstrate the effectiveness of the adapted methods in the actual application conditions, confirming the expediency of their use to increase the efficiency of image analysis in optical-electronic surveillance systems.