Google Street View Appears

Google launched Street View on May 25, 2007 with panoramic street-level imagery from five US cities: San Francisco, New York City, Las Vegas, Miami, and Denver. The feature integrated into Google Maps as an alternative to the top-down satellite/map view, allowing users to navigate a 360-degree photograph taken from street level at specific addresses. The initial imagery was captured using cars equipped with camera systems from Immersive Media — their Dodeca 2360 spherical camera array captured images in 11 directions simultaneously — combined with GPS, inertial measurement units, and laser rangefinders to record the precise position and orientation of each photograph. Processing the raw camera output into Street View panoramas required multiple pipeline stages: synchronizing images from multiple cameras taken at slightly different times, lens distortion correction, color calibration across cameras, photographic stitching into equirectangular panoramas (2D projections of the full sphere), GPS/IMU fusion to place each panorama at an accurate geographic coordinate, and compression for web delivery. The initial Street View cars were Chevrolet Cobalt sedans with the camera assembly mounted on the roof; Google subsequently developed its own Street View camera generations, reaching 15 cameras plus additional sensors by later versions.

Privacy concerns emerged almost immediately after launch. The initial release did not blur faces or license plates, and the panoramic resolution was sufficient to identify individuals captured walking on public streets, sunbathing in private yards partially visible from the street, or entering sensitive locations such as clinics or legal offices. After public pressure and requests from privacy advocates, Google announced in 2008 that it would apply automated face blurring and license plate blurring to existing and future imagery using computer vision — a significant engineering challenge given the volume of imagery. The same technology later became capable of blurring house numbers. International expansion brought region-specific privacy challenges: Street View launched in Australia, Japan, New Zealand, and several European countries in 2008-2009, and in Germany, widespread public opposition and legal action led to approximately 244,000 German households submitting opt-out requests to have their homes blurred from the imagery before German Street View launched in November 2010. Greece temporarily required Google to suspend Street View operations in 2009 over privacy compliance questions. In several countries, laws were passed or precedents set addressing whether photographing private spaces from public roads constituted privacy violations.

Street View’s technical capabilities and dataset grew substantially over the following years. The Street View Trekker, a camera backpack, enabled imagery collection in locations inaccessible to vehicles — hiking trails, university campuses, ski slopes, the Grand Canyon’s rim trails, and Amazon rainforest paths. Google Art Project (2011) and Street View for Business extended indoor 360-degree photography to museums, restaurants, and retail spaces. By 2019, Google reported that Street View covered over 5 million miles of roads across more than 80 countries. The accumulated dataset also proved valuable beyond navigation: Google used Street View imagery to train computer vision systems that could read house numbers (a problem Google researchers published about in 2014, noting their system read all house numbers in France in less than an hour with 98 percent accuracy, using the same convolutional neural network approach that had reduced word error rates in speech recognition), recognize street signs, identify changes to road infrastructure for map maintenance, and provide visual context for address geocoding. The technique of geolocating an image purely from its visual content (geolocation from appearance) became a research area building on Street View data, demonstrated prominently through the GeoGuessr game (launched 2013, showing random Street View panoramas and asking players to identify the location from visual cues) and eventually through neural network-based systems that could place images to within a few kilometers from the visual content alone. For autonomous vehicle development, Street View imagery — combined with HD mapping from dedicated survey vehicles — provided training data and localization reference maps that companies working on self-driving systems used alongside their own sensor data.