Abstract: Aiming at the high-precision and low-latency requirements of ultra wide band (UWB) radar for child presence detection (CPD) in automotive systems, this paper proposes a child presence detection algorithm based on deep learning one dimensional convolutional neural network (1DCNN). The algorithm introduces the Fire module to optimize computational efficiency and improve detection accuracy. Meanwhile, to address the issues of high near-range Tap noise and correlation between adjacent Taps in UWB signals, specific Tap signals are selected for downsampling processing, and temporal features are incorporated to further enhance the algorithm's accuracy and real-time performance. Experimental results show that the algorithm occupies less than 30 KB of storage space in a low-computational-power, microcontroller unit (MCU) environment, achieving a test accuracy of 99.9%, which is significantly superior to traditional methods. This algorithm has been applied to in-vehicle child presence detection in an automobile and successfully passed the China new car assessment program (C-NCAP) child presence detection scenario benchmark test.

Key words: UWB; 1DCNN; child presence detection; Fire module

摘要： 针对超宽带（UWB）雷达在车载系统里儿童遗忘提醒（CPD）时的高精度、低延迟需 求，文章提出了一种基于深度学习一维卷积神经网络（1DCNN）的儿童存在探测算法。该算 法通过引入 Fire 模块，优化了计算效率并提高了检测精度。同时，针对 UWB 信号中近距离 Tap 噪声较大和相邻 Tap 相关性的问题，选取特定 Tap 信号进行下采样处理，并结合时序特征， 进一步提升了算法的准确性和实时性。实验结果表明，该算法在低算力小的存储空间的微控 制器单元（MCU）环境下，占用空间不到 30 KB，测试准确率达到 99.9%，显著优于传统方法。 该算法应用于一款汽车中的车内儿童存在探测，顺利通过了中国新车评价规程（C-NCAP）的 车内儿童存在探测场景摸底测试。

关键词: UWB；1DCNN；儿童存在探测；Fire 模块