This paper addresses the challenge of reading low contrast text on tyre sidewall images of vehicles in motion. It presents first of its kind, a full scale industrial system which can read tyre codes when installed along driveways such as at gas stations or parking lots with vehicles driving under 10 mph. Tyre circularity is first detected using a circular Hough transform with dynamic radius detection. The detected tyre arches are then unwarped into rectangular patches. A cascade of convolutional neural network (CNN) classifiers is then applied for text recognition. Firstly, a novel proposal generator for the code localization is introduced by integrating convolutional layers producing HOG-like (Histogram of Oriented Gradients) features into a CNN. The proposals are then filtered using a deep network. After the code is localized, character detection and recognition are carried out using two separate deep CNNs. The results (accuracy, repeatability and efficiency) are impressive and show promise for the intended application.

Distance estimation is required for advanced driver assistance systems (ADAS) as well as self-driving cars. It is crucial for obstacle avoidance, tailgating detection and accident prevention. Currently, radars and lidars are primarily used for this purpose which are either expensive or offer poor resolution. Deep learning based depth or distance estimation techniques require huge amount of data and ensuring domain invariance is a challenge. Therefore, in this paper, we propose a single view geometric approach which is lightweight and uses geometric features of the road lane markings for distance estimation that integrates well with the lane and vehicle detection modules of an existing ADAS. Our system introduces novelty on two fronts: (1) it uses cross-ratios of lane boundaries to estimate horizon (2) it determines an Inverse Perspective Mapping (IPM) and camera height from a known lane width and the detected horizon. Distances of the vehicles on the road are then calculated by back projecting image point to a ray intersecting the reconstructed road plane. For evaluation, we used lidar data as ground truth and compare the performance of our algorithm with radar as well as the state-of-the-art deep learning based monocular depth prediction algorithms. The results on three public datasets (Kitti, nuScenes and Lyft level 5) showed that the proposed system maintains a consistent RMSE between 6.10 to 7.31. It outperforms other algorithms on two of the datasets while on KITTI it falls behind one (supervised) deep learning method. Furthermore, it is computationally inexpensive and is data-domain invariant.

Lane detection is of critical importance to both the self-driving cars as well as advanced driver assistance systems. While current methods use a range of features from low-level to deep features extracted from convolutional neural networks, they all suffer from the problem of occlusion and struggle to detect lanes with low or no evidence on the road. In this paper, we use a lane boundary marker network to detect keypoints along the lane boundaries. An inverse perspective mapping is estimated using road geometry which is then applied to the detected markers and lines/curves are fitted jointly on the rectified points. Finally, missing lane boundaries are predicted using lane geometry constraints i.e., equidistant and parallelism. Reciprocal weighted averaging ensures lane boundaries with strong evidence dominate their predicted alternatives. The results show a significant improvement of +7.8%, +6.8% and +1.2% of F1 scores over the state-of-the-art on CU-Lane, Caltech and TuSimple datasets, respectively. This proves our algorithm's robustness against both occluded and missing lanes cases. Furthermore, we also show that our algorithm can be combined with other lane detectors to improve their lane retrieval potential.

In dam inspection tasks, an underwater robot has to grab images while surveying the wall meanwhile maintaining a certain distance and relative orientation. This paper proposes the use of an MSIS (mechanically scanned imaging sonar) for relative positioning of a robot with respect to the wall. An imaging sonar gathers polar image scans from which depth images (range & bearing) are generated. Depth scans are first processed to extract a line corresponding to the wall (with the Hough transform), which is then tracked by means of an EKF (Extended Kalman Filter) using a static motion model and an implicit measurement equation associating the sensed points to the candidate line. The line estimate is referenced to the robot fixed frame and represented in polar coordinates (rho&thetas) which directly corresponds to the actual distance and relative orientation of the robot with respect to the wall. The proposed system has been tested in simulation as well as in water tank conditions.