Tree crown is an important index for tree species identification,estimation of tree DBH and forest volume,and monitoring tree growth. UAV remote sensing has the advantages of low cost and high precision,which is very suitable for obtaining high-resolution images. This paper took Metasequoia glyptostroboides forest in Qingshan Lake scenic spot,Lin'an District,Hangzhou as the research object,used UAV to obtain remote sensing images with different flight heights and different canopy density as the data source,and extracted the average crown width based on object-oriented method. Taking the average crown width measured on the ground as a reference,under the condition of relatively low canopy density,when the flight altitude were 65m,70m and 75m,the crown extraction accuracy were 95.39%,94.80% and 94.29% respectively. Under the condition of relatively high canopy density,the extraction accuracy of crown width were 89.10%,88.10% and 88.03% respectively. The results showed that within the research range,with the increase of UAV height,the accuracy of crown width extraction gradually decreased,and it was also applicable in two different canopy closures,which was more than 88% in two models. This method is efficient and reliable,and has important practical significance in forest resources investigation.