Low-Cost Embedded Stereo Depth Estimation Using Dual ESP32-CAM Modules for Robotics Applications

Stereo vision is a technique used by a machine to
calculate distance by comparing 2 images. These are all taken in
slightly different spots. In this paper we introduce our proposed
solution for low-cost stereo depth system with two ESP32- CAM
modules. One of the modules functions as an access point (AP).
The second module acts like a client. They are grayscale cameras
that produce 160 x 120 pixel images. This system implements a
block-oriented Sum of Absolute Differences (SAD) procedure for
disparity finding. Stereo triangulation is then used to estimate
depth. The implementation is based on fixed-point aritmethic,
and also it takes advantage of PSRAM to save memory usage
and reduce processing costs. It has the ability to create depth
maps and simple 3D point plots. The output of the skript is
an object coming 35cm right infront of the system. This test is
performed with static (non-moving) depth scenesOpen in new tab
i.e. the results also supports that if we have flat and low texture
areas then more noisy depth maps we get. In its configuration,
the system is a low-cost starter for depth sensing in robotics.