Research on real-time image edge detection algorithm based on multidirectional gradient and Otsu adaptive threshold

To overcome the limitations of the traditional Sobel edge detection algorithm—such as limited gradient directions and manually set thresholds that often cause wide edges and false detections—this paper proposes an improved method based on an eight-direction gradient Sobel operator combined with Otsu adaptive thresholding. Using MATLAB built-in grayscale images for simulation, the algorithm enhances texture recognition through multi-directional gradients and achieves automatic adaptive binarization. Experimental verification on MATLAB and ModelSim platforms shows that the PSNR decreases from 8.54 dB to 8.07 dB as Gaussian noise increases (σ = 0–25), indicating stable edge extraction and good robustness under noisy conditions. Compared with the traditional Sobel operator, the proposed method improves edge detection accuracy and reduces false edge rates by 8%–26%. Finally, FPGA-based real-time implementation with an OV5640 camera further confirms its practical effectiveness.