VirCap: Virtual Camera Exposure Control based on Image Photometric Synthesis for Visual SLAM Application

This is an official repository of

VirCap: Virtual Camera Exposure Control based on Image Photometric Synthesis for Visual SLAM Application, Shuyang Zhang, Jinhao He, Bowen Yang, Yilong Zhu, Jin Wu, Jianhao Jiao, Jie Yuan

This paper is still under review. The full paper and code will be released when the review process is finished.

Highlights

• A camera exposure control framework that employs photometric image synthesis techniques to attain update and control frequencies surpassing those of the camera’s hardware limitations.
• An exposure allocation method that balances image background noise and motion blur, using an intrinsic metric from a single image instead of interframe information.
• Sufficient experiments that shows the performance improvement in drastic illuminance change and motion blur case, on the backend application of visual SLAM.

Main Ideas

Virtual Capture

• Our VirCap does not need to interact with the camera's hardware interfaces to update the exposure parameters (exposure time and analogy gain) online; Thus, the camera's parameters can be updated at a very high frequency (~200 Hz).
• We use exposure bracketing technique to capture a group of images with different exposures to backup information over a high dynamic range.
• We implicitly maintain and update a target exposure, and the output images are generated by
  1. Selecting an appropriate base image from the bracket;
  2. Synthesizing a virtual image with the target exposure for output.
• An explaination from the imaging principle
  • All the images in the bracket do not introduce the analogy gain (set to 0 dB)
  • Our VirCap introduces image gain into the images after the imaging is finished; thus the gain process is peeled off from camera's hardware interaction and can be software updated at a high rate.

Exposure Allocation

• Purpose
  • To divide the target exposure value to the exposure time and analogy gain
  • To balance both image background noise and motion blur
• According to our bracketing capture mode, the exposure allocation task turns (1) to firstly select a basic image (choose the exposure time) and (2) then calculate the synthesis amplification (choose the gain).
• It should be noted that for a certain target exposure, there can be different combinations of exposure time and gain.
  • Suppose a target exposure of 20 ms, we can
    1. use base image with exposure of 1 ms and amplify 20 times;
    2. use base image with exposure of 5 ms and amplify 4 times;
    3. use base image with exposure of 20 ms and do not amplify;
    4. ...
  • All the combinations are reasonable，the difference is that when the exposure time is larger, the potential motion blur will be stronger, but background noise will be lower; vice versa.
• We design an image metric balancing both image motion blur and background noise, to select a proper base image from the bracket.

Image sequences in a drastic illuminance change scenario. Images with red circle indicate the first frame of illuminance change.

The comparison between traditional camera control framework and our VirCap.