Person re-identification (re-ID) involves recognizing and matching people over various camera perspectives within a video surveillance system. It consists of a significant challenge in matching individuals across various camera perspectives. While most of the current systems heavily depend on RGB appearance, which is ineffective in low-lighting conditions, yet for security reasons, surveillance often needs to continue in poor lighting. To address this limitation, cross-modal person re-identification has the potential to be effective in heterogeneous camera networks that utilize both depth and RGB cameras. Cross-modal person re-identification faces data discrepancy constraint because the nature of RGB and depth data is different. Existing work addresses this by segmenting the image into six horizontal strips to extract local shape information on both RGB and Depth modalities. However, when the individual is distant from the camera, some segments may not capture any part of the person's body, leading to inaccurate feature extraction. This study presents a technique for analyzing cropped full-body images using Histogram of Oriented Gradient (HOG). This method extracts local shape information and improves visual clarity regardless of lighting conditions. Additionally, we suggest an innovative metric learning strategy that enhances re-identification accuracy across RGB and depth modalities. This is achieved by comparing HOG features extracted from six horizontal image segments against those from the complete image. Experimental results from BIWI RGBD-ID and IAS-Lab RGBD-ID datasets confirm the effectiveness incorporated into our technique.