Lens: Gathers light and projects the scene onto the imaging medium’s surface. Some are single-lens, while others require multiple layers of glass for better imaging effects.

Filter: The visible light spectrum that the human eye perceives is limited, while the light spectrum that image sensors can identify is much broader. Therefore, filters are added to eliminate excess light wavelengths, allowing the image sensor to capture the actual scenes visible to the human eye.

Image CMOS sensor chip: This is the imaging medium that converts the image projected by the lens (light signals) into electrical signals.

Data processing circuit board: This transmits the electrical signals from the image sensor to the back end. For vehicle-mounted cameras, the circuit board will have more circuits, converting parallel camera signals into serial transmission for better interference resistance.

Workflow: Image input — Pre-processing — Feature extraction — Feature classification — Matching — Recognition completion

This involves inputting data from the camera and performing calculations such as detection, classification, and segmentation based on each frame of information, ultimately using multiple frames of information for target tracking and outputting relevant results;

1) Pre-processing includes frame formation, color adjustment, white balance, contrast equalization, and image rectification;

2) Feature extraction identifies feature points in the image based on pre-processing;

3) Object recognition is based on the output of feature data, classifying objects in the image — people, vehicles, traffic signs, etc., using machine learning, neural networks, and other algorithms.