Computer Vision Analysis & Monitoring of Vegetation Control Operations on Indiana Roadsides

Roadside vegetation control is an important part of modern infrastructure maintenance. Little data has been published to characterize the modern roadside mowing environment, identify challenges encountered by mowers, or quantify mower behaviors and responses to encountered conditions. A robust and effective methodology for collecting such information is also not rigorously defined in the literature. A better understanding of these factors, and more sophisticated tools with which to analyze them, may allow improving the efficiency, safety, and sustainability of the mowing process. In this document, it is shown that programmable action-sports cameras and standard agricultural telematics loggers can be used to record useful data about roadside mowing operations for understanding current operations and informing future developments. Image classification models detect human-operated mower incidents in the data gathered with the demonstrated equipment with rates of obstacle contact at up to 8 times per hour and of causing permanent damage to obstacles at up to 0.4 times per hour. The same model also shows human-operated mowers spend up to 60% of mowing time contacting road shoulders or lane surfaces on divided highways and up to 90% for rural roads. A visual odometry model was also tested on the gathered data to estimate tractor movement speed, but results from that model were inconclusive. The quantitative results reported here may enable future development of novel roadside vegetation control technologies, such as autonomous mowers or more advanced driver-assist systems for roadside tractor operations.