Real time image processing with IoT devices is a key enabler for autonomous systems like self-driving cars, drones and robotics. This work examines a new approach for image analysis on the smart devices which are restricted in computing resources when applied to IoT applications. The use of lightweight machine learning models such as YOLO is proposed in this system for object detection in real-time by relaying much of the processing to the edge to avoid relying solely on the cloud. A comparison of the results of object detection precision based on the image resolution shows higher accuracy when working with high-resolution images; however, more accurate images cause longer image analysis time, achieving 0.96 of precision at 1024*1024 pixels. It is also demonstrated how the inference time of the system falls in the range of 10–80 ms and how, when using simple models, the system operates faster at the expense of precision. In addition, the present observation of edge computing and cloud offloading clarifies that edge processing outperforms in the aspect of low latency for real-time applications where the processing load is small. They also show the possibility of performing object recognition in real time on IoT limited devices for the autonomous environment, which further facilitates the understanding of future improvements concerning precision, speed, and power consumption.