Table 3.

Comparative analysis of two kinds of object detection methods.

Methods	Onestage Object Detection	Twostage Object Detection
Principle	The input original image is processed directly to obtain the position coordinate value and category probability. The position is corrected thereafter.	Candidate regions are extracted from the input image through selective search and region generation network. Thereafter, convolution, pooling, and other processing is conducted to obtain feature maps.
Advantage	In the case of the low input separation rate, the speed and accuracy can be balanced simultaneously, and the detection speed is fast, which can reach above 45 FPS.	The deep semantic features of the object can be obtained. The detection accuracy of the object is high, whether it is a small object or a scene with considerable density.
Insufficient	Low accuracy for small objects and prone to miss detection, low positioning accuracy.	The algorithm has a large volume, large amount of stored data, complicated calculation process, and slow detection speed.
Realtime	Realtime.	Cannot reach real time