Adaptive HSV segmentation for real-time object detection under varying lighting conditions

Presented is a real-time object detection method based on the HSV color model, proposed as an efficient alternative to complex neural network models. The method is enhanced through the introduction of adaptive bounds, which makes it possible to account for local scene characteristics, including changes in illumination and background complexity. А comparative analysis was carried out between the proposed method and traditional algorithms as well as neural network models such as YOLOv8, using the Intersection over Union (IoU) accuracy metric. Experiments conducted on images with varying illumination levels and complex backgrounds confirmed the method’s ability to adapt to diverse conditions. The scientific novelty of the work lies in the development of an adaptive HSV-segmentation algorithm capable of dynamically adjusting color ranges based on local scene characteristics. This enables competitive accuracy under variable lighting and constrained computational resources—capabilities that have not been implemented in classical HSV segmentation methods. The "HSV segmentation" method demonstrated competitive results, especially under limited computational resources. At the medium tolerance level (the interval of permissible deviations of the HSV components from the central value) its accuracy reached 0.81 in terms of IoU, exceeding the performance of classical methods such as Canny Edge Detection and Template Matching and, in a number of cases, approaching the results of YOLOv8. The main advantages of the method are simplicity of implementation, low hardware requirements, and high processing speed, which makes it particularly useful for real-time applications. In addition, the method successfully provides "vision" for robots, solving tasks such as object detection, localization, and color-based recognition, as well as optimal trajectory calculation. This expands the possibilities for integrating visual systems into automated solutions, including resource-constrained platforms such as the Jetson Nano, and makes the method a promising tool for robotic applications.