Perception and Range Measurement of Sweeping Machinery Based on Enhanced YOLOv8 and Binocular Vision

As the “Green Concept” gains momentum, the state of road infrastructure has emerged as a topic of global concern. In parallel, the demand for road cleaning services provided by sweepers has experienced a dramatically increase. Consequently, the efficiency of road cleaning heavily depends on the capabilities of the sweeper, particularly its capacity for environmental perception and ranging. At present, the perception and ranging of road environments by sweepers is still largely rely on artificial observation, inefficient sensors, and traditional binocular ranging methods. These conventional techniques fall short in ensuring both cleanliness and driving safety of sweepers. This study introduces an enhancement to the YOLOv8 network, aiming to achieve precise environmental perception and ranging by integrating predictive frame resolution measurement with binocular stereo vision. Compared with the traditional binocular ranging method, the improve of binocular ranging via the YOLOv8 network effectively avoids the inaccuracies and misinterpretations stemming from incomplete parallax maps in traditional binocular ranging. This enhancement leads to heightened levels of accuracy and safety. Experimental results confirm that the enhanced ranging algorithm achieves an error rate of less than 0.5% under static testing conditions. Furthermore, the average error rate can be reduced to 0.78% during dynamic testing scenarios. Our ranging methodology significantly improves the precision of environmental road and distance data provided to sweepers in comparison to the pre-improvement binocular ranging detection. Post-improvement, the model retains its portability and versatility, making it well-suited for permanent integration. This model demonstrates notable migratability and generalisability.