Camera traps serve a pivotal role in the surveillance of wildlife populations within a given habitat. Nonetheless, commercially available camera traps tend to be relatively expensive, power-consuming, non-selective in their detection capabilities, and necessitate subsequent processing of the acquired images. In this study, we focus on developing a simple machine-learning wildlife camera using ESP32 CAM, a microcontroller that has an integrated camera, flash, and SD card storage, for detecting small mammals. The ESP32 CAM was designed to recognize and photograph squirrels in natural environments, employing an object detection model formulated via an accessible online machine-learning resource known as Edge Impulse. The efficacy of the models was systematically assessed, with training images derived from two distinct repositories, namely, the Google Images Repository and original photographs captured at the designated field site, compared for analysis. Model A, which utilized images from Google, achieved an accuracy rate of 61.5%, whereas Model B, which relied on local field photographs, attained an accuracy of 83.3%. We compared ESP32 CAM to Hawkray Digital trail camera to assess the model’s accuracy in detecting squirrels. Results from the Chi-Square analysis reveal that the number of squirrel pictures taken by both cameras was the same. A thorough discussion of various factors influencing the model’s accuracy, including background uniformity, size of the object, posture, and distance, was also undertaken. Furthermore, this study addressed the implementation of TinyML within the ESP32 CAM context, as well as the inherent limitations associated with the ESP32 CAM technology.