Image Processing Techniques for UAV Vision-Based River Floating Contaminant Detection

Unmanned Aerial Vehicles (UAVs) are becoming more and more widely employed in environmental inspection. The low-flying UAV carrying a High Definition (HD) camera is very helpful for river surface inspection since manual patrolling rivers is time-consuming and labor-intensive. To intelligently detect floating contaminant through the riverine images taken by UAVs, a series of image preprocessing work are investigated in this paper to deal with the specific problems of UAV-vision. Different from existing object detection approaches, image registration is carried out before Region of Interest (ROI) segmentation and contaminant detection to ensure the matching of images captured by the UAV at different times. According to the image structure, the statistical features of pixel values is used to achieve the extraction of the river region. The detection of floating contaminant is realized by means of anomaly detection based on the combination of Temporal Differencing and Background Subtraction method. The differencing process is improved in our work by utilizing the difference of image texture features instead of pixel-wise value. Experiments demonstrate that our approach can effectively detect the floating contaminant on river surface through riverine images taken by the UAV and is of good practicability.