City-Scale Visibility Graph Analysis via GPU-Accelerated HyperBall

Researchers Alex Hodge and Melissa Barrientos Trinanes have published a breakthrough paper introducing a system that massively accelerates Visibility Graph Analysis (VGA), a core space syntax method used by urban planners and architects to understand how spatial layouts influence human movement. The traditional bottleneck has been the all-pairs breadth-first search (BFS) computation, which limited practical analysis to small areas. Their novel approach combines three key innovations: delta-compressed CSR graph storage using LEB128 encoding for ~4x compression, the first application of the HyperBall probabilistic distance estimator (based on HyperLogLog counters) to visibility graphs to reduce algorithmic complexity, and custom CUDA kernels that stream and decode compressed data directly in GPU shared memory.

This fusion of techniques delivers unprecedented performance. Benchmarked against the industry-standard depthmapX tool using identical visibility algorithms, their system achieves a 239x end-to-end speedup when analyzing a graph of 42,705 cells. Crucially, it scales to previously intractable problems, processing a massive graph of 236,000 cells representing 4.8 billion edges in just 137 seconds. At a precision setting of p=10, their method's Visual Mean Depth calculation maintains a near-perfect correlation (Pearson r=0.999) with the exact BFS result, with a median relative error of only 1.7%. This means urban designers can now run accurate, city-wide visibility and accessibility simulations in minutes instead of days, unlocking new possibilities for data-driven urban design at a metropolitan scale.