Here is the video of how the Spy Object works.

Github Repository is here

When the program is run, camera takes three consecutive photographs. Every image is scanned to determine dominant color of every pixel. Pixels are converted into dominant color (either red , green or blue). Now, entire image is scanned to determine dominant color of the entire image. This color is printed out. If all three images are of a predefined color sequence, an audio file is played. If the sequence does not match, program returns nothing.

Lessons Learned–
1.The original idea was to to do face detection using RPi. We couldn’t find much precedence on that, also processing via RPi makes it very slow. The only method of doing it is creating a database of predetermined face (say 9 different expressions and angles) and train the Pi to detect the face using this database. This method is not scalable since if more faces are to be detected, larger database has to be built, which can not be handled in Pi.
2. We reduced the size of the image (160×120 pixels) to decrease the time it takes to process the image. Processing time is very high for images larger than that.
3. Color detection is not very accurate. We don’t know if it is the lights, reflection or the camera. Camera can detect the dominant color of a pixel(orange to pink are taken in as red and so on for blue and green) but differentiating between three closely related colors proved to be difficult. Possible solution here would be to print RGB value for a colored object and then manually determine a range of detection.