Design of an Embedded Binocular Stereo Vision Hardware and Software Platform Based on AM5728 DSP+ARM

Binocular Stereo Vision is an important form of machine vision. It has broad application prospects because it can easily calculate the depth information of objects relative to the camera within the camera's field of view using image information from the left and right cameras. Especially with its development over recent decades, stereo vision has been increasingly applied in fields such as robot vision, aerial surveying and mapping, reverse engineering, military applications, medical imaging, and industrial inspection. However, the technical complexity of binocular stereo vision technology means it is rarely applied in embedded scenarios. Due to its inherent large computational load, binocular stereo vision technology often requires implementation on PC platforms or through the collaborative efforts of DSPs and FPGAs.

This paper focuses on systematically researching the design of an embedded binocular stereo vision hardware and software platform, and delves into the principles of binocular stereo vision technology and stereo matching algorithms. The embedded binocular stereo vision platform designed in this paper uses the TI AM5728 DSP (C66X core) and ARM (ARM15 core) multi-core heterogeneous embedded processor as the main processor. The left and right cameras use two OV7670 image sensors. The entire platform uses a total of 3 AL422B FIFO chips to cache left and right images and the depth information after matching. The software primarily implements functions such as extraction of left and right camera images, stereo matching, distance measurement, and depth information output. The implementation process of the embedded binocular stereo vision platform in this paper can be briefly summarized as follows: 1. Design of the hardware platform's principles and physical implementation; 2. Calibration of the left and right cameras by capturing standard images to obtain the camera's intrinsic and extrinsic parameter matrices; 3. Integrating the obtained camera intrinsic and extrinsic parameters into the ranging algorithm, porting it to the AM5728 platform, and optimizing it; 4. Achieving target matching for left and right images, measuring the distance to a specific point in the image, and outputting the complete depth information image.