To aim the laser accurately at the target, we designed our device such that it has to align a crosshair-like box marked out in the middle of the frame over the center of the target before turning on the laser. As long as the middle of the bounding box surrounding the target is within this crosshair-like box, we take it as the device is centered on the target. That is where we had to make another crucial decision which involved the size of the crosshair. Generally, smaller crosshairs would result in greater accuracy since the device has to more precisely aim at the center of the target before the laser is engaged. However, the trade-off of doing so was that it resulted in a slight reduction in the responsiveness of the device. We found that with a small crosshair, the device kept attempting to recenter the bounding box before firing the laser at the object even though the object was stationary. This was due to the slight changes in the coordinates of the bounding boxes across each frame, which in turn kept changing the coordinates of its center. As the crosshair-like box was small, there was little leeway before the center of the bounding box went outside the boundaries crosshair-like box. Hence, the device kept on repositioning itself before firing the laser for every single frame which resulted in a slight reduction in responsiveness. Since the object detection algorithm we are using, MobileNetV2, is not good at detecting distant birds, we do not need to be extremely precise in centering the target and decided to have a larger crosshair.
