Spark Detection for Hot Work Based on HSV Color Features and Optical Flow Variations

Splashing sparks generated during hot work are a major cause of industrial and construction fires. Characterized by small scales, intense brightness mutations, and random motion trajectories, they pose significant challenges for real-time visual detection. To address the issues of high false detection and missed detection rates of traditional methods under complex lighting and background conditions, this paper proposes a spark detection method that fuses HSV color features with optical flow motion features. First, a YOLO-based adaptive threshold parameter optimization mechanism is introduced to dynamically optimize HSV color intervals and improve the stability of color segmentation. Second, a temporal motion analysis module is constructed by combining corner detection, pyramidal Lucas–Kanade optical flow tracking, and RANSAC-based global motion compensation, which is used to extract spark trajectories and suppress the impact of camera jitter. Finally, a fusion strategy based on Intersection over Union is adopted to integrate color and motion detection results, enabling spatial consistency judgment of spark regions. Experimental results show that the proposed method achieves an average Precision of 86.69%, a Recall of 75.90%, and an F1-score of 0.8156 on 8 groups of hot work videos, significantly outperforming baseline models such as the HSV fixed-threshold model, HSV autothreshold model, and pure optical flow model. Ablation experiments verified the effectiveness of each module. Overall, the proposed method exhibits excellent detection accuracy and robustness and can provide reliable technical support for intelligent fire hazard monitoring in hot work scenarios.