Abstract: Roadway shape control represents a critical aspect of roadway excavation. Photogrammetry, as a simple and efficient measurement technology, has been applied to perceive the contour formation of roadway cross-sections. However, due to the complex environmental characteristics of roadways and the limited space in the excavation face—where external equipment installation is impractical—noise and distortion inevitably occur during image acquisition via photogrammetry, thereby affecting measurement accuracy. To achieve simple and precise measurement of roadway contour dimensions and enable accurate roadway shape control, this study investigates the mapping relationship from the spatial coordinate system of excavation equipment to the imaging coordinate system under monocular vision. A measurement system is established with the near-field local coordinate system (NF-LCS) as the standard, and the advantages of NF-LCS over traditional coordinate systems are discussed. A control mechanism for real-time measurement of excavation contour formation based on NF-LCS is developed, accompanied by a regulation mechanism for different distances between equipment and the cross-section. To validate the proposed method, a dynamic visual perception experimental platform for contour parameters was constructed in a laboratory. Experimental results demonstrate that when cutting with 28, 30, and 32 mm drill bits, the measurement model based on NF-LCS yields a maximum dynamic measurement error of 1.4 mm for the linear dimensions of the cross-sectional cutting. In comparison, direct measurement based on projection relationships exhibits a maximum error of 5.6 mm. The use of a reference object reduces the maximum measurement error by approximately 67% for 28 mm and 30 mm drill bits, and by approximately 37.5% for the 32 mm drill bit, confirming significantly higher measurement accuracy with a reference object than without. Additionally, the study reveals the variation law of measurement error with the distance between the cutting cross-section and the reference object: with a fixed distance of 50 cm between the monocular camera and the reference object, the measurement error is minimized at 105 cm and maximized at 35 cm. The error decreases when the distance ranges from 5 cm to 25 cm and from 35 cm to 145 cm. Furthermore, the regulation law of measurement error in non-strictly parallel scenarios between the reference object and the cutting cross-section is summarized: the error increases nonlinearly with the angle between them. When the angle exceeds 40°, the error significantly increases, reaching a maximum of 35.787 mm. Within a 30° angle range, the error is comparable to that in the parallel state, and aligning one end of the contour to be measured with the camera's optical center effectively suppresses error growth. This visual perception method, requiring no external equipment or complex software-hardware systems, enables low-cost application in space-constrained and complex excavation faces. It effectively mitigates visual measurement errors caused by camera distortion in coal mine roadways and achieves simple, precise, and dynamic measurement of roadway contour linear parameters.