Our team is on the field for 360° street-level imagery capture; we are doing this initiative in partnership with HeiGIT gGmbH at Heidelberg University. So far, our field team has visited 37 wards, covering 1,300 kilometers of roads while navigating through all streets.
The captured imagery is being used for multiple applications. HeiGIT gGmbH at Heidelberg University focuses on detecting waste littering the streets, while at OMDTZ, we upload data to Mapillary, supporting navigation, providing local insights, and enabling automated data extraction. One of Mapillary’s key advantages is its ability to detect and generate data points such as traffic signs, amenities, fire hydrants, and other infrastructure. When these data points are uploaded into OpenStreetMap (OSM), they significantly enhance humanitarian efforts and urban planning.
Through this process, we see immense value in the efficiency of this data collection methodology. Using Bajaji (tricycles), field data collection can be done?in just one to three days per ward, with a few additional days for processing compared to the two weeks or more required for traditional door-to-door surveys.
?
Despite its efficiency, some challenges remain with the localization of the information being generated. We assume, based on training datasets, that Mapillary’s algorithm sometimes misclassifies objects such as:
?? ?? Green and blue tricycles (Bajajis) are mistaken for trash cans. https://bit.ly/438fHfR
??Street vendor tables are identified as benches. https://bit.ly/438fM39
?? Identifies street vendors and their products at Kariakoo Market as temporary barriers: https://lnkd.in/dDTYKizU
Still, we are looking to see if Mapillary could work with local communities to add training data so that the automation reflects specific regions, cultures, and settings. Its obvious observations show that biases in data point generation are largely influenced by the training data samples. For example, in Europe, trash cans are typically large, two-wheeled, and colored green or blue. In contrast, in Tanzania and many African countries, a similar description refers to tricycles (Bajaj). Another case is at Kariakoo Market, where vendor tables are intentionally placed along roadsides. While Mapillary identifies these tables as barriers—which is correct in some contexts—this interpretation is inaccurate for Kariakoo, where the tables serve a deliberate trading purpose.
#data #omdtz #mapillary #bajaj
