Human Detection and Its Distance Measurement in Agricultural Fields by Stereo Image Processing

Having an alarm system to detect human can reduce risk of accidents and human injuries in agricultural environments. The purpose of this study was to develop a stereo vision system for human detection and distance estimation by agricultural vehicles. Asphalt road, harvested wheat and sugarcane fields were selected for imaging in three human gestures, three sunlight conditions, and intervals of 3–21 m. The histogram of oriented gradients (HOG) algorithm and support vector machine (SVM) classification method were applied to identify humans using MATLAB image processing toolbox. The human detection rate (HDR) and total accuracy indices were used for the algorithm evaluation. The HDR was dependent on the human gesture and image background complexity. The highest HDR, equal to 98.96%, was obtained in standing back to the camera gesture. The average HDRs on asphalt road, wheat, and sugarcane fields were 97.95, 89.11, and 91.55%, and the average accuracy were 97.51, 92.33, and 92.88%, respectively. As the human–camera distance increased, the HDR and accuracy decreased in all three environments. The maximum and minimum HDR values were 98.63% and 79.23% that occurred at distances of 9 m and 21 m on asphalt road and wheat field, respectively. The minimum and maximum values of the accuracy were 95.13% and 97.64% that occurred in 21 m and 3 m distances, respectively. In total, the proposed algorithm performance was acceptable up to 12 m distance, but it needs to be optimized for increasing the HDR and accuracy in far distances. © 2023, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.