Rotational Odometry using Ultra Low Resolution Thermal Cameras

Abstract

This letter provides what is, to the best of our knowledge, a first study on the applicability of ultra-low-resolution thermal cameras for providing rotational odometry measurements to navigational devices such as rovers and drones. Our use of an ultra-low-resolution thermal camera instead of other modalities such as an RGB camera is motivated by its robustness to lighting conditions, while being one order of magnitude less cost-expensive compared to higher-resolution thermal cameras. After setting up a custom data acquisition system and acquiring thermal camera data together with its associated rotational speed label, we train a small 4-layer Convolutional Neural Network (CNN) for regressing the rotational speed from the thermal data. Experiments and ablation studies are conducted for determining the impact of thermal camera resolution and the number of successive frames on the CNN estimation precision. Finally, our novel dataset for the study of low-resolution thermal odometry is openly released with the hope of benefiting future research.

Figures (4)

• Figure 1: Data acquisition setup. The $24\times 32$ thermal camera is connected to a readout board which translates its I2C interface to a serial interface via USB. A $100 \mu$F decoupling capacitor is used for providing a stable power supply to the thermal camera. The thermal camera is mounted on top of a servo motor controlled by a micro-controller via serial interface over USB. This setup enables the acquisition of thermal camera data while rotating the camera at precisely-controlled speeds.
• Figure 2: CNN architecture for the estimation of rotational speed from thermal camera data. The CNN is composed of two convolutional layers (with max pooling in between), followed by two fully-connected layers and an linear output layer. This small-size architecture has been designed with the aim of reducing the CNN compute complexity for potential implementation in CNN accelerator hardware googlecoraltpu.
• Figure 3: Box plot of the 6-fold test MSE in function of the number of consecutive thermal input frames $N_f$. The red line indicates the median value. The best $\text{MSE}_{\text{test}}$ is achieved for $N_f=3$.
• Figure 4: Box plot of the 6-fold test MSE in function of the thermal camera resolution subsampling factor $N_r$. The red line indicates the median value. As expected, the lower the thermal image resolution, the higher the $\text{MSE}_{\text{test}}$.