Prevention of loss of muscle functions is important for the elderly because loss of muscle functions is accompanied by decreases in quantity and quality of exercise, resulting in lowered capability in activities of daily living and progression of frailty. In general, muscle strength is an index of physical capacity. However, current methods of measuring muscle strength of the lower limb are limited because they require large-scale measuring equipment. Therefore, easy measurement of muscle strength using home-based method is desired. Some research has suggested easy muscle force evaluation methods by correlating muscle strength and walking velocity based on easy methods of muscle strength evaluation such as 10-m walking time and timed up and go test. However, these walking tests are not natural movements in daily life. In addition, these tests require a well trained assessor. The assessor has to provide reproducible and objective results. Considering all of the above-mentioned issues and needs, we propose a highly accurate method for estimating muscle strength in the lower limb measured during walking while wearing sensors in the ankle. From the input data of standard walking waveform obtained from a walking subject, the muscle strength was estimated using multiple linear regression analysis and convolutional neural network (CNN). The mean absolute errors in multiple linear regression analysis and CNN were 17.3% and 9.27%, respectively. Therefore, we confirm that the CNN model provides highly accurate estimation of muscle strength in the lower limb.