Diagnosing spatial mismatch between objective and street-view-based perceived walkability assessments: A dual-track framework for street-renewal diagnostics

Street walkability has both objective and perceived dimensions, yet the extent, spatial distribution, and robustness of mismatches between these assessments remain underexamined. Using central Nanjing, we developed a dual-track framework. The objective track integrated street-network, point-of-interest (POI), and street-view imagery data and used the Criteria Importance Through Intercriteria Correlation (CRITIC) method to construct an objective walkability score (OWS). The perception track used Qwen3-VL-Plus, a multimodal large language model (MLLM), to evaluate street-view images across six dimensions (safety, lively, beautiful, wealthy, depressing, and boring), producing a perceived walkability score (PWS). Model ratings were compared with human ratings for 500 images. Agreement was estimated on those same images, with no independent validation set, and may therefore be optimistic. Pearson correlations across the six dimensions ranged from 0.37 to 0.71. The original-scale composite PWS had an intraclass correlation coefficient of 0.60 and a mean absolute error of 0.85. High-value areas of OWS and PWS did not coincide, and the standardised perception gap score (PGS) remained spatially clustered at both the sampling-point and street-segment levels (Moran's I = 0.306 and 0.247, respectively; permutation p = 0.001 for both). Exploratory geographically weighted regression (GWR) estimated spatial variation in associations between built-environment indicators and PWS. Relative OWS and PWS levels classified sampling points into four diagnostic types, two of them defined by divergence between the assessments. By quantifying and locating the spatial mismatch between objective and perceived walkability, the framework provides diagnostic evidence for incorporating visual perception into fine-grained street-renewal assessment.