With the rapid development of low-altitude unmanned aerial vehicle technology, small consumer-grade UAVs equipped with optical sensors can quickly and flexibly acquire high-resolution image data of target objects, which presents a broad prospect for expanding applications in various fields of geology. UAV-SfM is the latest technological method for imaging 3D stereo construction that is a core technology for deepening research in the field of low-altitude UAV technology geology, but at present the lack of research on the comprehensive accuracy of data results obtained by using the UAV-SfM method has affected the further promotion and application of this technical method. This paper addresses whether the DJI Phantom 4 RTK consumer-grade UAV has the technical potential to be used for detecting shallow surface changes in the mountainous areas of the central Yunnan plateau, and selects a typical sloping field in the red land of Dongchuan as the test area. The DSM and DOM data from the same area were obtained using the key SfM-MVS technique. In order to evaluate the accuracy of the repeated UAV observations on typical sloping land, 3D point accuracy evaluation was carried out for the bare sloping land I and the growing sloping land Ⅱ in the experimental area, using 3D discrete point sampling based on profile lines and 3D point set sampling based on window surfaces, respectively. The point accuracy analysis showed that: (1) In 3D discrete point sampling and precision analysis based on profile line, the mean precision error of slope farmland Ⅰ plane point position is ±0.029 m, and the accuracy of 3D point position error is ±0.072 m. The mean accuracy of slope farmland Ⅱ plane point position is ±0.032 m, and the 3D point position error is ±0.075 m. (2) Based on the 3D point set sampling and precision analysis of the window surface, the mean precision error of slope farmland I plane point position is ±0.013 m, and the 3D point position error accuracy is ±0.066 m. The mean precision error of the slope farmland Ⅱ plane point position is ±0.038 m, and that of the D point position is ±0.076 m. The comprehensive analysis shows that the evaluation accuracy of single point sampling based on profile line is better than that of 3D point set sampling based on the window, but the plane accuracy and vertical accuracy can reach centimeter level on the whole. According to the experimental comparative analysis, different surface roughness impacts on the UAV repeated observation accuracy, and the 3D point position error with large surface roughness is larger than that with small surface roughness. The research results of this paper can provide a quantitative reference for the precision control and acquisition scheme setting of geomorphic data acquisition and 3D reconstruction based on UAV and SfM methods.