This paper presents the design and implementation of an integrated human activity recognition and safety monitoring system tailored for indoor rescue personnel. The system combines a waist-mounted inertial measurement unit, a foot-mounted LiDAR sensor, and a barometric pressure sensor to enable real-time monitoring of movement, activity, and altitude. A three-stage signal processing pipeline refines multi-sensor data to enhance step detection and inter-limb distance estimation, while lightweight machine learning models deployed on embedded microcontrollers accurately classify seven essential actions and eight movement directions. Experimental evaluations show that the activity recognition module achieves an F1-macro score of 98.43% with a compact 796 kB XGBoost model, and the movement direction module reaches 100% accuracy using a decision tree classifier. The system further demonstrates reliable fall detection, accurate altitude estimation under challenging indoor conditions, and positioning accuracy consistently below 2 m during linear trajectories. The proposed framework advances infrastructure-free indoor positioning by addressing cumulative drift, environmental dependency, and resource constraints, thereby supporting safer and more effective rescue operations.