<p>There are two inherent problems in real-time gesture recognition: low accuracy and slow recognition time, as well as a limited number of recognizable gestures, and gesture recognition is affected by illumination, leading to misjudgment. To tackle these issues, a gesture detection and segmentation method based on multifeature extraction is proposed. The model includes a gesture feature extraction module and a real-time gesture recognition and segmentation module. The gesture feature extraction module uses Xception and ResNet50 models to extract features from the same input image, combines these features, and fuses them with contour features and histogram of oriented gradients (HOG) features to comprehensively extract gesture features. The real-time gesture recognition module uses the camera to capture the gesture area and improves the segmentation accuracy by applying a skin color segmentation method. In addition, the model also uses an explicit attention mechanism to ensure accurate gesture capture. The experimental results show that this method significantly improves the accuracy of gesture recognition and performs well under different illumination conditions. Only a few misjudgments occur under extremely weak or bright light conditions, with fast response and strong robustness.</p>