While 3D vision technology was still being debated among structured light, monocular/binocular vision, and ToF solutions, the team firmly chose the line laser + binocular vision technology route. This approach improved the 3D vision's light resistance to 180,000 lumens, achieving an accuracy of 0.02mm and extending the depth of field to a range of 500-4000mm. 3D vision is no longer limited to industrial assembly line scenarios with fixed ambient light, truly achieving indoor and outdoor versatility, strong light resistance, anti-reflection capabilities, and full-scenario adaptability in dim environments. Line laser + binocular vision 3D technology is rapidly penetrating traditional manufacturing, building a "new paradigm" for industrial digital transformation.
Agricultural Machinery Innovation: Self-developed 3D Stereo Camera
At that time, Xinjiang's cotton production had ranked first in the country for 22 consecutive years, but the labor costs for harvesting each year amounted to 15 billion RMB. When evaluating market potential, the team pointed out that if they could capture 10% of the cotton harvesting robot market, this single item alone could conservatively generate 1.5 billion RMB in revenue.
However, a thorny problem arose during the development of agricultural harvesting robots. Although the coordinate robot had been successfully developed, the lack of a reliable visual positioning system severely affected the recognition accuracy and harvesting efficiency. The team had also purchased 3D vision solutions from foreign brands. However, without exception, these vision systems could not fundamentally meet the actual needs of cotton harvesting. They could neither solve the problem of recognizing cotton nor adapt to the working requirements of high-intensity outdoor light.
Currently, the mainstream 3D stereo vision technologies on the market mainly include structured light solutions, time-of-flight (ToF) solutions, and binocular vision solutions. Although they differ in application scenarios and scope, without exception, regardless of which technology solution is chosen, there are certain limitations.
Structured light works on the principle of laser speckle encoding, using an active measurement method, usually consisting of a laser or DLP and a pair of binocular cameras. Laser point measurement illuminates the working surface with hundreds of thousands of laser points, and the left and right binocular cameras build xyz coordinates to generate a structured image. DLP emits a light source of a specific wavelength, and the light source with encoded information is projected onto the object. The distortion of the returned encoded pattern is calculated through algorithms to obtain the object's position and depth information. Currently, most companies on the market generally use structured light solutions. However, this solution faces a series of challenges in outdoor environments with strong sunlight:
1) Intense sunlight can reduce the visibility of laser points, making it difficult for the binocular camera to accurately capture the position information of the laser points.
2) In situations with a large depth of field, light interference may prevent the left and right binocular cameras from clearly distinguishing the laser points, leading to image aliasing and affecting measurement accuracy.
3) It is highly sensitive to external light sources and may be interfered with by other light sources in the surrounding environment, affecting the accuracy of depth measurement.
4) The depth of field is generally narrow, and the field of view is limited. Due to its susceptibility to light, it can usually only operate under constant indoor lighting conditions.
The ToF (Time-of-Flight) technology works by continuously sending light pulses to the target and then using a sensor to receive the light reflected from the object. The distance to the target is determined by detecting the flight time (round trip) of these emitted and received light pulses. The sensor calculates the distance of the scene being photographed by calculating the time difference or phase difference between the emitted and reflected light, thereby generating depth information. Combined with traditional camera imaging, this allows the three-dimensional contour of the object to be presented as a topographic map with different colors representing different distances.
However, although the ToF technology is low-cost and has a fast response speed, its imaging is limited by the pixel resolution of the structured light pattern. The imaging surface of the structured light pattern is usually limited to 600,000 pixels, and the ToF solution often falls short in terms of imaging accuracy and detail capture. Therefore, the ToF technology is limited in high-precision, large-scale industrial applications and is more commonly used in consumer applications with relatively lower accuracy requirements.
In contrast, the visible light binocular stereo vision method works by observing the same object from two viewpoints to obtain images of the same object from different perspectives. The three-dimensional image of the object is obtained by calculating the positional deviation (disparity) between image pixels using the principle of triangulation. Its advantages include high resolution, high accuracy, high resistance to strong light, and low cost.