Detecting suspicious behaviors is important for surveillance and monitoring systems. In this paper, we investigate suspicious activity detection in thermal infrared imagery, where human motion can be easily detected from the background regardless of the lighting conditions and colors of the human clothing and surfaces. We use locally adaptive regression kernels (LARK) as patch descriptors, which capture the underlying local structure of the data exceedingly well, even in the presence of significant distortions. Patch descriptors are generated for each query patch and for each database patch. A statistical approach is used to match the query activity with the database to make the decision of suspicious activity. Human activity videos in different condition such as, walking, running, carrying a gun, crawling, and carrying backpack in different terrains were acquired using thermal infrared camera. These videos are used for training and performance evaluation of the algorithm. Experimental results show that the proposed approach achieves good performance in suspicious activity recognition.