This study presents the results of a human detection algorithm using a single camera installed in front of
the tractor. An algorithm was designed and implemented in five stages named DSWTS (Division,
Segmentation, Watershed techniques, Thresholding, Subtraction). The algorithm first changes an input
RGB image to a grayscale image. This image is then divided into small blocks. So, deletion of some
unneeded regions makes image processing more comfortable. To segment an object from the background,
the edges of the object are detected using a magnitude gradient function and watershed techniques.
Then, by subtracting human from background and comparing series of image frames, the pedestrian is
recognized. The algorithm was evaluated under morning, noon and evening lighting conditions. Its
results were compared with the histogram of oriented gradient (HOG) method and the cascade method
that are commonly used to identify humans in images. The results show that the DSWTS algorithm has
good accuracy at 8–20 m. Also, in order to improve its performance from 0 to 8 m distances, it combined
with two other algorithms. Then by comparing and evaluating combined algorithms, best fusion was discovered
and found good results.