The majority of road accidents are caused by human oversight. Advanced Driving Assistance System (ADAS) has the potential to reduce human error and improve road safety. With the rising demand for safety and comfortable driving experience, ADAS functions have become an important feature when car manufacturers developing new models. ADAS requires high accuracy and robustness in the perception system. Camera and radar are often combined to create a fusion result because the sensors have their own advantages and drawbacks. Cameras are susceptible to bad weather and poor lighting condition and radar has low resolution and can be affected by metal debris on the road.
Clustering radar targets into objects and determine whether radar targets are valid objects are challenging tasks. In the literature, rule-based and thresholding methods have been proposed to filter out stationary objects and objects with low reflection power. However, static vehicles could be missed and thus result in low detection accuracy. To overcome these drawbacks, a data-driven method has been proposed, which uses a variety of features and thus is more suitable for complex real-world scenarios.
Data-driven methods require a large amount of labeled data. In this paper, we propose a data-driven radar object detection and clustering method aid by camera data. As cameras have high accuracy in object detection, it is used to train a classifier to determine whether radar object is valid. The algorithm is validated with real-world driving data and has shown good performance in object detection.