In this document, we discuss the use of optical filters in stereo vision 3D cameras. More specifically, we are going to first investigate the optical characteristics of different filter types and then optimize the 3D camera performance under challenging lighting conditions with properly selected optical filters, where significantly improved imaging quality can be observed in test images.
1. Introduction
The performance of stereo vision 3D cameras is directly related to the quality of the input stereo image pairs. Therefore, in certain scenarios, it’s beneficial to use additional optical components to enhance the imaging quality of the stereo camera, thereby improving the performance of the depth perception. In this document, we focus primarily on optical filters, which can avoid critical imaging issues if applied appropriately and are generally avalaible at affordrable cost.
2. What is an Optical Filter?
An optical filter is an optical component designed to selectively transmit or block specific wavelengths of light. They are typically made of specific materials or coated with films to achieve desired optical properties, used to control the propagation of light in optical systems. For stereo vision 3D cameras, placing filters of specific wavelength bands on top of the cover glass can help selectively transmit or block light within certain wavelength ranges, thereby can significantly enhance imaging quality.
3. Types of Optical Filters
3.1 Neutral Density Filters:
These filters are designed to reduce the intensity of incident light. This filtering action is non-selective, meaning it reduces the intensity of light for all wavelengths equally, thereby decreasing the transmitted light intensity without affecting the spectral distribution of the light. In photography, they are commonly used to prevent saturation in high dynamic range images. While in laser applications, they are typically used for controlling laser power.
3.2 Broadband filter:
These filters selectively transmit a certain wavelength band of light while blocking other wavelengths. They are further classified into short-pass, band-pass, and long-pass filters, each allowing specific wavelength ranges to pass through while attenuating others. Different bandpass filters are chosen based on the desired transmittance characteristic over the spectra of interest. The following diagram illustrates the spectral distribution curves of different types of bandpass filters.
a. Band-pass filter：This filter selects a specific wavelength band to pass through while attenuating wavelengths outside the passband. The spectral curve is depicted in Figure 1.
Figure 1 Spectral Curve of Wide-pass Bandpass Filter
b. Short-Pass: This filter allows light shorter than the selected wavelength to pass through while blocking light longer than that wavelength. Common examples include visible light pass filters, as shown in Figure 2.
Figure 2 Spectral Curve of Visible Light Pass Filter
c. Long-Pass : This filter allows light of wavelengths longer than the selected value to pass through while blocking light of wavelengths shorter than the selected value. Typical examples are IR Pass filters with visible light cut-off, as shown in Figure 3.
Figure 3 Spectral Curve of IR Pass Filter
3.3 Band-Stop Filter:
This filter blocks only a specific wavelength band of light and is essentially the opposite of a bandpass filter. It is commonly used to eliminate light of a particular wavelength and plays an important role in scientific applications such as fluorescence microscopy imaging.
3.4 Polarizing Filter:
This type of filters is used to adjust the polarization state of incident light. Common types of polarized light include linearly polarized light, circular polarized light, and elliptical polarized light. It is used in applications such as laser polarization imaging, liquid crystal displays, and sunglasses.
4. Challenges of Stereo Vision 3D Cameras
4.1 Sensitivity to Ambient Light:
Stereo vision usually captures image pairs with ambient light in the environment. However, differences in light intensity and incident light angles between the left and right cameras can lead to significant differences in brightness between the left and right images of a stereo pair, resulting in a sharp decline in algorithmic matching effectiveness.
4.2 Poor Performance in Monotonous and Textureless Scenes:
Stereo vision relies on visual features for image matching. Therefore, scenes lacking visual features, such as the sky or white walls, can pose matching difficulties, leading to significant matching failures or even errors. Adding laser speckle projection can enhance features in low-texture scenes, but in certain situations, ambient light can reduce the contrast of laser speckles or even obscure their grayscale information.
5. Using Filters
5.1 Filtering out Ambient Light
The majority of ambient light in indoor scenes consists of visible light, with its spectrum concentrated in the 400nm-600nm range. For stereo vision 3D cameras, areas with high brightness or highly reflective objects can result in incorrect depth calculations due to excessive ambient light. Adding an IR pass filter effectively suppresses overexposed areas caused by ambient light, ensuring the accuracy of depth calculation. Figures 4 and 5 show a comparison of before and after adding the IR pass filter. By adding a filter, the depth performance of the ground has significantly improved.
Figure 4 Infrared and depth images without an IR pass filter
Figure 5 Infrared and depth images with an IR pass filter
5.2 Enhancing Speckle Signal-to-Noise Ratio (SNR)
The signal-to-noise ratio (SNR) of laser speckles in stereo vision 3D cameras has great impacts on imaging quality. Ambient light introduces background noise to laser speckles, resulting in poor SNR or even overshadowing of speckle brightness. Adding an IR pass filter to eliminate visible light effectively improves the speckle SNR, especially in low-texture scenes, thereby enhancing imaging quality. Figures 6 and 7 show a comparison of before and after adding the IR pass filter. By adding a filter, By adding a filter, the depth performance of the light tube and the cart has significantly improved.
Figure 6 Infrared and depth images without an IR pass filter
Figure 7 Infrared and depth images with an IR pass filter
