Vision-Based Moving Obstacle Detection and Tracking in Paddy Field Using Improved Yolov3 and Deep SORT

• DOI: 10.3390/s20154082
• Publication Date: 2020-07-23T15:26:01Z
• Authors: Zhengjun Qiu; Nan Zhao; Lei Zhou; Mengcen Wang; Liangliang Yang; Hui Fang; Yong He; Yufei Liu
• Keywords: Tracking (education); RGB color model; Machine Vision

Using intelligent agricultural machines in paddy fields has received great attention. An obstacle avoidance system is required with the development of machines. In order to make more intelligent, detecting and tracking obstacles, especially moving obstacles fields, basis avoidance. To achieve this goal, a red, green blue (RGB) camera computer were used build machine vision system, mounted on transplanter. A method that combined improved You Only Look Once version 3 (Yolov3) deep Simple Online Realtime Tracking (deep SORT) was detect track typical figure out center point positions fields. The Yolov3 23 residual blocks upsamples only once, new loss calculation functions. Results showed obtained mean intersection over union (mIoU) score 0.779 27.3% faster processing speed than standard self-created test dataset (human water buffalo) acceptable performance for could be real field an average 5-7 frames per second (FPS), which satisfies actual work demands. future research, proposed support agriculture flexible autonomous navigation.