Preserving mobility, the ability to keep a correct posture and dynamic balance in order to walk properly, is fundamental to maintain autonomy in daily life. Based on the correlation between muscle groups and autonomy, previous research has suggested that maintaining muscular tone in knee extensors is critical. Continuous training of knee extensors during aging is therefore essential to maintain independence. In this work, it is hypothesized that it is possible to estimate knee extensor activity only from IMU data based on a simple lower limbs model. The accuracy of the knee extensor activity estimation algorithm has been tested using sEMG measurements as control data on three different walking patterns: normal walk, fast walk and stair climbing. Estimated knee torque area and measured muscular activity for each step were compared confirming a high estimation accuracy with a correlation efficient R=0.80. Moreover, muscular activity can be divided based on intensity in three groups of statistically significant difference confirmed by the Steel-Dwass method. Future works should test the usability of the algorithm for different walking patterns, and use the collected data and the refined algorithm to implement a smart resistive device to increase knee extensor exertion during each walking pattern to the level necessary for sufficient extensor training.