In both high-performance computing and Internet of Things (IoT) applications, it is important for programmers to utilize the hardware as efficiently as possible in terms of both performance and power usage. Automatic methods of tuning code for such efficiency have been developed and used for many years in HPC, but less so in other areas. In this paper, we apply these methods to the domain of IoT to explore the feasibility of using autotuning in this domain. Our case study examines the application of roadway traffic analysis. We use Computer Vision techniques to count the cyclists, pedestrians, and cars as they travel through a video of a campus intersection. A signifficant portion of the tracking computations is the Kalman filter method, which is used to estimate the paths of objects through the video. This Kalman filter is the target of this autotuning study. We present results for the performance and power usage of the system with multiple autotuning techniques and optimized BLAS libraries on three disparate architectures. Our results show that autotuning is both feasible on and beneficial to IoT platforms.