| 
                            			 
                              			
                              		 | 
                            		
                            		
																						Vehicle Automatic Detection Algorithm from UAV Videos
Based on Shape Analysis
											                            			
                            			
						
                            			 
                            				PENG Bo1,2,3, CAI Xiaoyu1,2, ZHOU Tao4, LI Shaobo2, ZHANG Youjie2, DUAN Lianfei5
                            			 
                              			2019, 38(04): 
																					15-22. 
																														DOI: 10.3969/j.issn.1674-0696.2019.04.03
																				
                              			 
                              			
                                		
			                            	In order to collect continuous traffic flow information correctly and comprehensively from a regional 
perspective, a vehicle automatic detection method was proposed based on shape analysis aiming at UAV (unmanned aerial 
vehicle) videos. Firstly, a ROI (region of interest) was marked manually on the video frame, and grayscale processing was 
conducted on the frame. Secondly, a sub-pixel skeleton image was generated based on Canny edge detection result of ROI, 
and the image skeleton was decomposed and reconstructed. Then, vehicle targets were recognized through comprehensive 
application of morphological operations (dilation, erosion, filling and closing) and connected components shape features 
(area, rectangularity, major axis and minor axis of equivalent ellipse). Finally, algorithm detection and manual 
inspection were respectively conducted on 548 UAV video frames, and correct detection rate, repeated detection rate, 
missed detection rate and false detection rate for vehicle detection were calculated. Test results show that the proposed 
algorithm achieves higher correct detection rate (averaging 95.02%), lower repeated detection rate (averaging 2.20%), 
missed detection rate (averaging 2.77%) and false detection rate (averaging 8.24%). Besides, standard deviations of the 
correct detection rate, repeated detection rate, missed detection rate and false detection rate are 2.09%, 1.67%, 1.67% 
and 2.56%, respectively, which indicates that the proposed algorithm obtains smaller discrete degree of performance 
indexes and higher stability.
			                             
                              			
                              					                              					                              					                              					References | 
	                              					                              					                              				Related Articles | 
	                              				Metrics
                              				                              			
         								                              		 |