At the heart of the patented technology is a baseline imaging approach that creates a visual reference of the vehicle’s underside. When the car is switched off and in a stable state, the system captures and stores an image of the underbody, establishing a snapshot of what the area should normally look like.

Later, when the vehicle is preparing to move or performing a safety check, new real-time images are taken and compared with the stored reference. Rather than processing the entire underbody scene every time, the software focuses only on areas where it detects a change from the baseline image. 

These altered sections are isolated as target images and sent for more detailed analysis, allowing the system to reduce computing demands while quickly identifying anything unusual beneath the vehicle.

AI safety technology could help prevent underbody accidents

Rather than repeatedly processing every component beneath the car, the patented system is designed to ignore elements that remain unchanged from one scan to the next. Fixed structures such as the battery pack housing, suspension components, aerodynamic covers, and other underbody hardware are treated as part of the reference environment, allowing the software to concentrate on anything that appears or moves after the baseline image has been recorded.

Once a new or altered area has been identified, the technology performs a more detailed examination of that specific section. It extracts visual features from the target image, analyses them using its detection algorithms, and determines whether the object is a living organism. The system can then assess the detected subject’s condition or status before deciding whether further action or a safety warning is required, CarNewsChina reported.

Unlike in-cabin monitoring technologies, identifying living organisms beneath a vehicle presents a far more complex computer-vision challenge. The space underneath a parked car is highly unpredictable, with constantly changing environmental conditions that can interfere with accurate detection.

Shifting shadows, uneven lighting, road debris, accumulated dirt, and different ground textures all create visual noise that makes reliable recognition difficult. Conventional motion-detection systems can also generate false positives, often failing to distinguish between ordinary environmental changes and the presence or movement of a person or animal beneath the vehicle.

Two-stage detection system could make parked cars safer

Instead of depending exclusively on object-recognition software, BYD’s patented technology first builds a unique visual reference of the area beneath each parked vehicle. By comparing new images against this stored baseline, the system can quickly identify anything that has changed in the underbody environment before carrying out a more advanced analysis.

Only after these differences have been isolated does the software apply its recognition algorithms to determine whether the detected object is a person, an animal, or another living organism. This layered, two-step process is designed to improve the accuracy of the detection system while cutting down on false alarms triggered by fixed structures or other non-threatening objects that remain stationary beneath the vehicle.

The patent also aligns with BYD’s broader push into advanced vehicle safety systems. The company recently unveiled another invention designed to detect forgotten occupants inside a car using radar-based sensing and signal analysis.

Together, the two technologies monitor opposite sides of the vehicle – one focusing on the cabin and the other on the space underneath. The approach suggests BYD is building a wider sensing ecosystem that combines computer vision, radar, and intelligent monitoring to improve overall vehicle awareness.

Recommended Articles