Infant fall detection is critical for the timely identification of fall events, assessment of the severity, and reduction of potential injuries. Traditional fall detection technologies typically rely on devices such as cameras, force sensors, accelerometers, and gyroscopes. While these devices provide accurate measurements, they are often expensive, require complex installation, and depend on external power sources, leading to higher system complexity and maintenance costs. This paper reports a self-powered, wearable sensor based on triboelectric nanogenerator (TENG) for infant fall detection, featuring a bridge-structured PDMS layer and a copper foil electrode. By attaching the TENG sensor to the skin or joints of an infant model or human body, we successfully detected fall status, frequency, impact intensity, and impact localization. The sensor achieves a minimum detectable acceleration of 0.4 g on human skin, with a sensitivity of 2.6 V/g. When integrated with artificial intelligence algorithms, the system achieves over 94% accuracy in predicting fall locations. Furthermore, the sensor maintains a stable output after tens of thousands of cycles, demonstrating exceptional stability and repeatability. Compared to traditional fall detection technologies, the proposed system offers several advantages, including low cost, simple manufacturing, easy installation, self-powered operation, and high portability.