Abstract: To address instability in navigation path recognition in potato plastic-covered planting environments, caused by strong plastic mulch film reflection, complex variations in natural illumination, and vibrations from machinery operations, a navigation line detection method has been proposed that integrated depth information with color image features．This method combined with Kalman filtering to achieve temporal smoothing of path parameters．This method first established a dual-channel（depth-color）parallel recognition framework．Depth image was restored by filling gaps using linear interpolation, and depth information was transformed into height distribution features．Geometric centers of ridges were then extracted using a row-wise regional adaptive threshold segmentation method．Color channel, an adaptive brightness normalization and color feature enhancement algorithm was developed to address complex field illumination conditions, including strong sunlight, backlighting, and cloudy conditions．This algorithm performed feature suppression, enhancement, and fusion within HSI color space to accurately obtain plastic mulch texture center．Subsequently, an initial navigation line was obtained based on a decision-level fusion strategy, and a Kalman filtering was introduced to predict and correct navigation line parameters across consecutive frames, thereby reducing path angle fluctuations caused by machine vibration and image noise．Field experiments have shown that under cloudy, sunny front-light, and sunny backlight conditions, recognition success rates of depth image method were 100%, 95.5%, and 98.5%, respectively, while those of color image method were 99.5%, 93.5%, and 96.0%, respectively．After dual-channel fusion, navigation line recognition success rate remained stable at over 97% under all working conditions．Kalman filtering reduced standard deviation of navigation angle deviation by 57.1% and the maximum angular jump by 74.1%．Overall method achieved navigation path angle deviation of less than 1°, with a single-frame processing time of less than 0.30 s, thereby meeting requirements for navigation accuracy, operational stability, and real-time performance of potato automatic seedling-releasing machine．An effective approach for visual navigation of agricultural equipment was provided for plastic-covered planting scenarios．