This Project researches techniques and algorithms for the detection, recognition and analysis of low-level and high-level features from mobile mapping data along urban and rural transport corridors.
Mobile mapping systems use integrated vehicle-borne positioning, imaging and laser scanning sensors to record data that cannot be readily acquired from aerial and satellite platforms. Examples include vertical building facades and street furniture. Low-level features include edges, corners, planar surfaces, and high-level features include buildings, road signs and the attributes of other man-made and natural features within the transport corridors.
The research uses fused imagery, video, LiDAR, GNSS and sensor orientation data recorded by mobile mapping platforms to recognise, locate, map and measure the spatial attributes and characteristics of features of interest.
Automatic feature extraction for spatial information product generation is essential for organisations such as local government authorities and other government agencies, as well as for 43pl companies, to more efficiently monitor and manage transport corridor infrastructure and associated assets such as power lines, roads and railways, and street-side furniture.
A workshop focusing on Program 2 research was held in Perth to discuss the requirements now and in the future for feature extraction from aerial and mobile laser and image data. The workshop was attended by 30 end users including representatives from Landgate, AAM, Whelans and Main Roads WA. The feedback gathered will inform research utilisation and future strategy. End-user interest centred on mobile mapping.
A new distance transform algorithm has been developed that performs as fast as other state-of-the-art algorithms, but is more stable being less variable in run-time performance with changing image content. The algorithm is generally faster than other state-of-the-art algorithms for images having sparse foreground points. Formal evaluation is yet to be undertaken, but preliminary results are promising. The algorithm is novel and has wide applicability to the field of computer vision and pattern recognition in general.
Two methods have been developed to be deployed in Landgate to aid in the processing of Earthmine data.