Abstract: To address the contradiction between limited computational resources and high performance demands faced by intelligent IoT devices in real-time vision tasks within edge computing scenarios, where traditional feature detection algorithms such as SIFT (Scale-Invariant Feature Transform) and SURF (Speeded Up Robust Features) suffer from high computational complexity and poor real-time performance, while lightweight algorithms like ORB (Oriented FAST and Rotated BRIEF) and BRISK (Binary Robust Invariant Scalable Keypoints) exhibit insufficient feature representation and robust matching capabilities in complex dynamic scenes, this paper proposes an entropy-guided multi-scale feature detection and matching method, termed EWPFDM (Entropy-Guided Wavelet Pyramid Feature Detection and Matching). The method constructs a wavelet pyramid-based multi-scale feature representation model and introduces a local information entropy adaptive threshold mechanism to achieve intelligent filtering of feature points. Meanwhile, the BRISK descriptor is improved to enhance rotation and scale invariance of the extracted features. In the matching stage, a hybrid metric strategy integrating entropy similarity and Hamming distance is designed to balance matching accuracy and computational efficiency. Experimental results demonstrate that the proposed method achieves an average inlier rate of 80.2% on the Hpatches standard dataset, outperforming existing algorithms by 6%–20%. Under challenging conditions such as illumination variations and viewpoint changes (θ = ±45°), it still maintains a correct matching rate above 70%. In FPGA implementation, the method occupies only 21.3% of hardware resources with a processing time of 52 ms per frame, which is comparable to the resource consumption of ORB, thereby exhibiting both superior matching performance and practical deployability in edge environments.