Visual Scheme—Contour Extraction


    The two cores of visual detection are "image + algorithm". Only after the image is acquired, the algorithm engineer can perform algorithm processing based on the acquired image information. Therefore, image acquisition is very important in visual detection. As the quality of the image acquired depends entirely on the lighting engineer, the lighting engineer plays a very important role in the early stage of the project. When high-quality or good images are acquired for the algorithm engineer to process, a lot of workload will be reduced in the algorithm processing process, and the processing results of the image will be given in time to determine whether the project can be continued.

     Then what is a good image and what is a bad image? First of all, the target object under test should be clearly distinguished from the background beyond the target object. Secondly, the target object to be photographed should be clearly visible, that is, the focus should reach the best focus position. Finally, the information points on the surface of the target object under test should be clearly distinguished from the target object. This will test the ability of the lighting engineer to select the preliminary scheme, such as detecting the size of each tooth of the outer teeth of the gear and the size of the inner circular teeth.

    Prime lens + backlight: adjust the brightness of the backlight, so that the background gray scale can reach 255, and the background and the gear can be clearly distinguished. Due to the inherent field angle of prime lens, the entire side of the inner circular teeth is fully visible, resulting in unclear contour of each tooth, difficulty for the algorithm to measure the size of each inner circular tooth, and low accuracy. At the same time, the outer teeth and the inner circular teeth are not in a plane, a focus position cannot be reached, and the images acquired are always blurred on one side and clear on the other side. Therefore, the image belongs to the inferior image.

Prime lens+ backlight

Telecentric lens + backlight: The image has been significantly improved. Both the outer teeth and inner circular teeth can be clearly distinguished from the background. The telecentric lens does not have the inherent field angle of prime lens, so that the contour of inner circular teeth is not affected by the side. Although the outer teeth and the inner circular teeth are not in a plane, the strong depth of field feature of the telecentric lens can reach a focus position in the depth of field and obtain a clear image. However, there is a certain difference between the gray scale of the outer teeth and the inner circular teeth in this image, which brings some troubles to the algorithm processing. Therefore, this image belongs to the good image.

Telecentric lens + backlight

Telecentric lens + parallel light source: There is no gray scale difference in the whole gear. The whole image sharpness is not disturbed by external stray light. The contrast is far better than other images. Therefore, it belongs to the high-quality image.

Telecentric lens + parallel light source

Simple algorithm is used to process each image:

Telecentric lens + parallel light source: The effective area of the image (inner and outer teeth) is completely covered.
Telecentric lens + backlight: A small part at the boundary of the effective area (inner and outer teeth) of the image is not covered.
Prime lens+ backlight: More parts of the effective area (inner and outer teeth) of the image are not covered. More algorithm processing is required for complete covering

The image effective area (internal and external teeth) is still not completely covered under more algorithm processing.
Simple algorithm processing on the images acquired with the three schemes can intuitively show the quality of the acquired images. Therefore, several feasible schemes shall be designed to collect better images. Then, each scheme shall be tested, and the acquired images shall be fed back to the algorithm in time. The final scheme shall be determined according to the algorithm processing results.