Real-Time HSV Transparent Object Shape Extraction and Projection

The existing mapping strategy based on Lidar and camera cannot support transparent object avoidance that is hard to recognize. To overcome this, we present a real-time and reliable detection approach to project transparent obstacle shapes in true size into map. To distinguish transparent object from environment, an extraction algorithm is proposed. Object shapes and sizes are kept in high accuracy by processing the extracted image in hue, saturation, and value channels to refine and merge candidate regions with different value and saturation under continuity constraints. We then build a projection map of transparent objects to facilitate obstacle avoidance. The size and shape comparison experiments in a series of pitch angle at different distances are conducted as the supplementary of angle selection to ensure high shape and size accuracy. Experiments indicate that proposed methods enable robot to conduct map construction and transparent obstacle avoidance in real-time.