When analyzing images in the far-infrared spectrum, the problem of separation of objects arises on the given temperature ranges. The simple threshold separation is hindered by the presence of a noise component. This task is most relevant when used to on weak computing systems and organizing data processing onboard the device (or mobile device). This problem may arise during data analysis: during processing onboard aircraft (UAV), automated control systems for robotic systems, security and access control systems, etc. In article proposes an approach that allows solving the problem of separating a thermal imaging image to objects within specified ranges. The approach is step-by-step image processing using the filtering and blurring method based on multicriteria image preprocessing. A method for changing the bitrate and a method for detecting objects based on the density of similar pixels in a given analysis sliding window. The filtering and blurring method is adapted and accelerated for implementation on low-performance computing modules. As a result of developing the method, we managed to lower the computational requirements. However, this slightly increases the time of preprocessing. The article describes in detail the algorithm for changing the multicriterial method, shows of changes in characteristics and their influence on the result. A method is proposed for changing the gradation of IR image ranges, based on multiple changes on the section gradient. The on a step-by-step merging algorithm for adjacent ranges is based on the use of adaptive weighting factors. The coefficients are determined automatically and depend on the frame. The method of final separation into objects is based on an analysis in the sliding window, we analysis of the number of different elements and the allocation of close objects in groups. The methods and algorithms proposed in the work are adapted for calculations on devices with reduced requirements for computational costs. On the set of test images obtained by portable thermal imaging cameras (SEEK with a resolution of 320 by 240 pixels) from the UAV, car, and street, we present the effectiveness of the application of the approach introduced in the article.