Contemporary technologies for point-type fire detection have made significant strides in mitigating the risks of financial losses, injuries, and fatalities caused by fires. This proposed system aims to harness advanced digital video processing and computer vision techniques to create a more efficient flame recognition system. In this study, we devised a flame segmentation algorithm with low complexity and high accuracy using dual detection segmentation (DDS). The process begins with the detection of fire regions based on the YCbCr color space model, which is refined through a thresholding algorithm. Once the fire and flame contents are segmented, key features are extracted using GLCM texture analysis and Oriented Rotated BRIEF features. These extracted features are then classified using the fast computation ensemble learning method of the Optimized RUSBoost Decision Tree (ORUS-BDT) algorithm, which combines Decision Tree and Boosting methods for ensemble classification of flame frame features to detect flames and determine their direction. The proposed method is evaluated through simulations, and Key Performance Indices (KPIs) such as accuracy, error, precision, recall, F1-score, correlation, False Positive Rate (FPR), and Kappa coefficients are calculated for comparison with previous implementations. The accuracy achieved by the proposed method is 98.5%, exceeding existing methodologies by 2–4%