Google Research has quietly made a breakthrough in the emerging area of image recognition and rapid video analysis — a breakthrough that has significant implications for pedestrian detection. Pedestrian detection refers to the analysis of statistics of pedestrian scale, occlusion and location. It gives detection systems a base of knowledge under which to operate.
Now on to Google's breakthrough.
As an experiment [PDF], Joe Yue-Hei Ng, Matthew Hausknecht, Sudheendra Vijayanarasimhan, Oriol Vinyals, Rajat Monga and George Toderici — researchers from the University of Maryland at College Park, the University of Texas at Austin and Google — applied several deep neural network architectures to analyze a dataset of videos over longer time periods than previously attempted.
It is a departure from the usual method of image-recognition analysis, which is convolutional neural networks, or CNNs.
The results were impressive, not just to the researchers ("our best networks exhibit significant performance improvements over previously published results…") but also to observers of this and related developments.
"At 15 frames per second, the Google team set a dramatic record that does not sacrifice either speed or accuracy against benchmarks like the Caltech Pedestrian detection metric," wrote Nicole Hemsoth at The Platform (hat tip to Hemsoth for first highlighting the study).
Pedestrian detection and prediction
The goals of the research are admirable and, OK, have the potential to be highly profitable. On that subject, Hemsoth writes that "use cases for how GPUs will power real-time services off the web are still developing. Pedestrian detection is one of those areas where, when powered by truly accurate and real-time capabilities, could mean an entirely new wave of potential services around surveillance, traffic systems, driverless cars, and beyond."
But as commenters to published accounts of this research project and other technologies deployed for pedestrian detection have noted, these emerging capabilities have scary implications as well.
First, though, a use case we can all get behind.
Driverless cars and pedestrians
It is widely acknowledged that advances in pedestrian detection will make driverless cars even safer, and also those driven by humans. However, there are limits to what the technology can do.
The California Institute of Technology, which established among the first and widely recognized datasets of such statistics, wrote in a paper [PDF] on the subject that "in the US, nearly 5,000 of the 35,000 annual traffic crash fatalities involve pedestrians — hence the considerable interest in building automated vision systems for detecting pedestrians."
It tested 16 "pre-trained state-of-the-art detectors across six datasets" on the market, and to make a long story short, found that "even though progress has been made, there is still much room for improvement."
Sign up for CIO Asia eNewsletters.