Abstract: Real-time and accurately acquisition the navigation centerline is often required for the interrow tea harvesting machine. In this study, an interrow navigation line algorithm was proposed to extract the tea ridge roads using tracking growth ROI and machine vision. Firstly, the fixed ROI (region of interest) was used, where the lower left region of the image was selected as the first ROI. The grayscaling was carried out in the ROI. The same ROI was grayscaled using a variety of RGB component combinations. A comparison was performed on the super green index grayscaling, as the best grayscaling. Secondly, the maximum class variance was used to segment the tea ridge road and background. The morphological operations were then carried out to reduce the noise. The coordinate information and feature points were extracted from the ROI, according to the maximum connectivity domain. The ROI was dynamically generated to extract all the information of the tea ridge road in the whole image. Finally, the least square method was used to obtain the trans-radial tea plucking machine chassis travelling between the ridge navigation line. A comparison was made on the accuracy and time consumption of the navigation line extraction, including the diffuse filling, dynamic ROI and the tracking growth ROI algorithm. The results show that the deviation of the heading angle after extraction was all in the range of -4°~4°. The absolute value of the deviation was greater than 4° in the heading angle of only two frames extracted by the dynamic ROI algorithm. The reason was that the larger convolution kernel of the image preprocessing was resulted in the loss of image information from some of the tea ridge roads. The ROIs were accounted for 34.47% of the total pixels of the image. The amount of the pixel computation was reduced by 11.82% on average, compared with the dynamic ROI. The average computation time of the single-frame images was 41.6 and 156.3ms lower than that of the dynamic ROI and diffuse filling algorithm, respectively. The errors in the variance of the navigation were 0.039° and 0.012° lower than those of the diffuse filling and the automatic ROI algorithm, respectively. The continuous frame and field tests were performed to verify the extraction. The navigation centerline of 1782 consecutive image frames in the video stream was extracted in the continuous frame test. The maximum magnitude of the extracted heading angle deviation was 2.41°, with an average absolute heading and standard deviation of 0.405° and 0.463°, respectively. The processing time of each frame was 44~14 ms, with an average processing time of 18 ms; The field test was conducted to collect the images of the tea rows and then extract the navigation line. The industrial camera (RER-USB4kCAM01H) was placed at the position of 0.2 m directly above the center of the front bracket of the cross-row tea harvester, with a tilt angle of -12°. The navigation angle was obtained -3.11° and the navigation error was 0.5°. The interrow navigation line extraction of the tracking growth ROI tea ridge was realized using navigation angle extraction under weed and leaf fall interference, fully meeting the requirements of the navigation center line extraction accuracy for the interrow interrow tea harvesting machine. The finding can provide a research basis to further study the unmanned system of tea harvester.