Publications / 2020 Proceedings of the 37th ISARC, Kitakyushu, Japan
We have developed a function to process the images of three monocular cameras mounted on a hydraulic excavator in real time, and to notify the user of any human-like images on the monitor and by sound. This function is composed of the following four functions. (1) Detection of the human head by random forest, (2) Normalization of the region of interest, (3)Evaluation of human presence based on luminance gradient features, and (4) Tracking based on the human and the excavator motion model. The algorithm is implemented in a FPGA unit, and the 3-camera images are input into the FPGA unit and processed by time-division manner to achieve both high-speed processing and low-cost.