Gait serves as an effective biometric for long-distance identification, particularly in scenarios where other biometric techniques present limited results. Most of the current gait recognition research relies on gait videos captured in controlled settings, predominantly using RGB cameras, while only a minority employ infrared cameras. There is a notable demand for real-time gait recognition in uncontrolled environments, especially utilizing infrared cameras for security and surveillance purposes. This study introduces a multi-frequency gait database constructed from long, medium, and short wavelength infrared (LWIR, MWIR, and SWIR) as well as visible (RGB) cameras in uncontrolled outdoor settings. The database encompasses recordings of individuals engaged in four distinct activities: normal walking, walking with a coat, carrying a backpack, and holding a briefcase. Additionally, it uses a knowledge-based system for silhouette extraction in dynamic environments. This research evaluates the robustness of state-of-the-art gait recognition methods to changes in environmental conditions, clothing, and carrying covariates by utilizing our dataset to establish a benchmark for databases captured across various frequency bands. Furthermore, it assesses gait recognition performance at lower scales (up to 0.05).