According to a recent study, every nine days a child dies from vehicular heatstroke in the US. It often happens when a child is left behind in the vehicle unintentionally due to in-attention or other reasons. Temperatures inside a locked vehicle are known to rise rapidly, sometimes leading to that fatal heatstroke.
Vehicle manufacturers have already identified this as a major safety problem that needs to be combated. Euro NCAP road map 2025 has also identified “child presence detection” as a tertiary safety system proposed to be mandated in future cars. Driven by the nature of the problem and the legislative push, multiple solutions have started evolving. This article discusses various sensing technologies that vehicle manufacturers are considering to address this problem, makes a case on why radar could be one of the suitable technologies and then goes on to explain how one can design a radar-based in-cabin monitoring solution.
Sensing Technologies for In-Cabin Monitoring
Child presence detection or in-cabin monitoring can be done with the help of a radar sensor or a camera or an ultrasonic sensor. A camera as a sensor for in-cabin monitoring (and mounted on the dashboard) is effective only for driver monitoring, but not for the entire cabin/car. Kids or pets present in the rear seat are not covered by a camera-based system as it is difficult to see through the seats and other passengers, and the camera also brings in the issue of privacy.
Ultrasonic sensors are cost-effective as well as reliable for an automotive application like park assist. However, the response rate for detection of objects present in the field of view is low. Classification of objects to human and non-human is also not possible with an ultrasonic sensor.
The alternative being radar, which has seen numerous automotive applications, catering to ADAS (advanced driver assistance systems) and AD (automated driving) as explained here. Radar is not only economical but it offers exceptional accuracy and reliability. A radar sensor mounted on the roof of the car can sense the cabin, detect objects as well as classify them with high accuracy.
How mmWave Radar Works for Cabin monitoring
Radar has numerous applications in cars in the form of ADAS, wherein sensors placed outside the car assist the driver in understanding the surrounding clearly. For cabin monitoring, a radar sensor is mounted on the roof of the car facing the seats. This, coupled with a calibrated field of view, helps with the detection of any live objects in the car- in this case, kids and pets. Having a radar sensor that does simple object detection is not enough here; one needs to have a supporting algorithm that can clearly distinguish/ classify live objects with others in the car. Radar-based cabin monitoring can be broadly classified into three steps:
- Object detection
- Human/Non-human classification
- Adult/child classification
A typical radar-based system will have a transmitter, waveguides, antenna,- receiver and a processing unit with necessary algorithm package. A sample algorithm pipeline for this application is depicted below.
Range and Doppler estimation: The ability of the monitoring system to differentiate two closely placed objects. Fast Fourier Transform is the preferred algorithm for Range and Doppler estimation.
Threshold: Filtering out noise points or irrelevant objects inside the cabin is called thresholding, which is generally done with the help of CFAR (constant false alarm rate) algorithm.
Direction of Arrival: The direction at which the reflected radar signal arrives back to the sensor is termed the direction of arrival and to detect this, an array of sensors is required. Usually, the sensor apparatus hosts a beamforming technique to roughly estimate the direction of the signal. ESPRIT is the preferred algorithm for DoA.
Zone-based Grouping: Grouping of reflected points for each object is termed as clustering or grouping. In Zone based clustering, zones are predefined in cars as a result reflected points in this region are grouped together.
Feature Extraction: A dense group of point cloud data is converted to a more manageable form for further processing. Micro Doppler is the preferred algorithm for feature extraction application.
Classification: For an application like in-cabin monitoring, it is critical for a radar device to differentiate or classify living and non-living things. This helps in differentiating something like a child seat with an actual kid. SVM is the preferred algorithm for classification.
In-cabin monitoring for child presence detection is one of the key safety features required in vehicles. Radar, as the sensing technology, fits the application requirements while also curtailing the privacy concerns that other sensing technologies such cameras have. If you are looking to develop radar-based in-cabin monitoring solutions, do reach out to us!