Why Infrared Images?
With the fast pace of development in deep learning, object detection has gained a huge progress. Among all of the different objects, living objects are the most important ones in autonomous driving.
However, in some real-world scenarios, object detection with RGB cameras fail to generate reasonable detection results. Moreover, we found that it is beneficial to use infrared cameras on detecting living objects according to heat emission.
Here, we list a few cases we found in FLIR dataset which infrared camera help to improve object detections.
- Day scene
In the figure above, we can see that the people on the side walk are not detected in RGB image, yet we can easily get detection results with infrared image.
- Night scene
Night scene object detection is more challenging than day scene for RGB cameras due to insufficient illumination. Nevertheless, with infrared camera, we can detect people more easily.
- High dynamic range (HDR) scenes
During driving, it is inevitable to encounter with high dynamic range scenes. For example, driving out from a tunnel/building, driving in dust, … etc. In the case of driving in dust shown as above, RGB camera can hardly detect objects while thermal camera has reasonable detection results.
- Motion blur
Motion blur is commonly seen with RGB cameras, especially in night scene for having longer exposure time. Yet, this problem does not exist with infrared cameras. As you can see, the image with motion blur does not detect the car and the person in the middle while infrared camera can detect them both.
- Detecting small objects
We also found that it is more difficult to detect small object with RGB cameras, but we can get detection results with infrared cameras.