The results for obstacle detection are measured by setting the distance where the obstacle should stop. If the car is told to stop when it sees an obstacle let’s say 40 cm, it shall stop when it receives data that says there is an obstacle at a 20cm range. Not all objects can be detected in the range we set. Some have to be closer to the sensor in order to allow the sensor to detect it, for example skinny objects may not get hit by the Sharp IR Range Sensor beam and cause the sensor not able to receive the data. It sometimes has to go closer than the set range to the skinny or thin object in order to be able to detect it.
The results for traffic light detection are measure by detecting the traffic lights from different range. We measured how much seconds the program fails to detect the traffic lights in one minute. If the traffic light is detected 60 seconds out of the 60 seconds, we marked its success rate as 100%, if the traffic light is detected for a total of let’s say 58 seconds, we mark its success rate as 58/60 or 97%(rounded). Most of the time when the program fails to detect only lasts for one or two seconds, which probably wouldn’t be a big deal. The green lights are a lot dimmer compared to the other lights.
The lane detection processing speed varies from about 100 milliseconds to 150 milliseconds, mostly about 110 milliseconds. The lane detection managed to keep the processing speed below 333 milliseconds which will not cause delay for the webcam to receive the next frame. With the traffic light detection added, it would be an average total of 210 milliseconds per frame, and a maximum of 300 milliseconds per frame which is still under 333 milliseconds, so it would still not cause any delay from receiving frame from the camera. The lane detection’s time is the time required to go through 3 lane models that we made, each lane model about 90 cm in length, with a total of 270 cm. The stopwatch is stopped once the car has reached the last lane.
The face detection processing speed varies from about 120 milliseconds to 140 milliseconds, mostly about 110 milliseconds. Our webcam used in this project is 30 frames per second, meaning that there each frame is 333 milliseconds. The traffic light detection managed to keep the processing speed below 333 milliseconds which will not cause delay for the webcam to receive the next frame.
detection is measured out of 30 seconds. Within 30 seconds, we figured out of
how many seconds the computer failed to detect the face. The face detection failed to detect
the face mostly because of the distance and glasses, but this will not really
effect the driver in the real world because we believe that the driver may not
be about one meter away from the steering wheel when he or she is driving the
car so that failed data caused by distance may not be a big problem. Glasses
may also cause false detection, which we may try to fix in the future