Visual place recognition is crucial for camera localization and loop closure detection. This paper presents SCHAL-Net (Synergistic Constraint of Holistic and Local Features Network), a dual-stream cross-modal model addressing RGB-infrared image recognition challenges in complex lighting. Unlike conventional pre-trained backbone methods, our approach improves robustness through synergistic global and local feature constraints. The network incorporates shallow feature enhancement for inter-modal correlations and cross-modal high-dimensional mapping for discriminative global representations. KAIST dataset evaluations show significant performance gains, with 13.1% and 36.3% improvements in Top-1 and Top-10 metrics respectively, achieving 22.4% and 65.5% state-of-the-art results.