*5.1. Computation Complexity Analysis*

The average time for one backpropagation (BP) iteration per frame is ~19.7 ms with the aforementioned computer implementation. Considering a full BP run with the maximum of 100 iterations over the train set of 60 (for examples), the maximum training time is 60 × 100 × 19.7 = 1.97 min. The total time for a forward pass of a single image to obtain the score vector, which corresponds to the system latency is about 1.57 ms. A system with a latency of 50 ms or less, is considered a real-time system. However, the proposed system operates at a speed that is higher than 30x the real-time speed. The average frame rate that was practically achieved was ~99 frames per sec (fps). This is a significant increase in the frame rate comparing other works reported in the literature.
