assignment

individual assignment: write an application that interfaces a user with an input and/or output device that you made

group assignment: compare as many tool options as possible

[1] As for my final project I wanted to create a way to measure activity in urban public parks, I decided to create an application or interface that would take as an input the ESP32-Cam video stream and use real-time object detection to detect different objects or activities that the ESP32-Cam would observe in urban parks. I continued with the lessons and work from the previous week. I used the networked video stream I achieved in Week 11 as the input of a program that would then take that video stream and use YOLO Real Time object detection to process images at 30 FPS using a pre-trained model that would be able to detect a limited number of objects. This proved to be much more difficult than I imagined and I had to get the help of Amira to figure out why the code I had written wasn’t working. We figured out the problem was that I hadn’t installed the correct pre-trained weight file for YOLO, so my program wasn’t able to detect anything at all.

Once I correctly installed the pre-trained weight file, my locally run program was able to detect objects that were streamed from the ESP-32 Cam. With this I then created a list and a dictionary to store the items and add them to a running list that would be able to keep track of the objects the program had “seem”. This would allow me to measure activity in different places, including urban parks. I decided to create two additional ESP-32 Cams to deploy in different parks for a given time, and see if I could compare and contrast the activity that the Real Time object detection would be able to measure in different places.

One problem I encountered was that I hadn’t realized that the Real Time Object detection I was using wasn’t able to differentiate if it had already “counted” or “seen” a previous object or person. So, it would count a person or object multiple times, 1 per observation in a frame. As video has multiple frames, one same person who walks by the ESP32 Cam, would be counted multiple times. In the future, I plan to continue exploring how to fix this problem. For now, I was content with the results, and instead decided to adapt what I could measure. Instead of stating I was able to measure the number of people/objects, I would only be able to measure the relative time an object/person was in the camera’s line of sight. This would still be able to provide interesting insights, especially when comparing different deployments.

Download Files