Used OpenCV to access the webcamera and create a video feed along with threading to allow for requesting OpenAI API responses parallel which identify the objects in the frame passed along with their positions. The bse_prompt.txt file allows the program to be flexible and capable of handling generalized object detection.
My default base prompt is 'people or objects', but one could tailor the program to detect just cars, etc. One could also adapt the program to keep track of counts or other metrics for the objects detected.
The openai_request_buffer can be used to limit the frequency of the OpenAI requests, by requesting the next response in openai_request_buffer seconds.